具体实施方式

下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。

以下，术语“第一”、“第二”仅用于描述目的，而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括一个或者更多个该特征。在本公开实施例的描述中，除非另有说明，“多个”的含义是两个或两个以上。另外，“基于”或“根据”的使用意味着开放和包容性，因为“基于”或“根据”一个或多个所述条件或值的过程、步骤、计算或其他动作在实践中可以基于额外条件或超出所述的值。

图1为本发明实施例提供的一种车道参考线的提取方法的流程图之一，该方法可以包含如下步骤：

步骤101、获取路面点云数据，所述路面点云数据是采集车在行驶过程中采集的路面上的点云数据，所述路面上包括多条车道线。

高精度的地图采集车在路面上行驶过程中，采集整个路面的点云数据，得到路面点云数据。该路面点云数据中包括路面上的一切对象的点云信息，如路面、路面上的车道线、路面上的异物等。

路面点云数据中的每个点包括的信息有：三维坐标、采集时间戳、颜色信息信息（RGB值）、反射强度信息（IQteQsity）。

步骤102、从所述路面点云数据中获取起始点和根节点，所述起始点和所述根节点为人工预确认的车道线上的两个点；所述起始点到所述根节点的方向与所述采集车的行车方向一致；所述人工预确认的车道线为所述多条车道线中的其中一条。

起始点和根节点用于作为生成车道参考线（下文简称为参考线）的最初两个点。此处的起始点和根节点是在人工预确认的车道线上提取的，因此，本方案最终生成的参考线是车道线中的其中一条。

本方案中的参考线是生成路面上其他车道线的参考标识，具体用于作为生成其他车道线的方向参考和长度参考。在得到参考线之后，通过复制平移的方法可以得到路面上的其他车道线。

相关技术中通过全部人工点选的方式得到一条车道线，将该车道线作为生成其他车道线的参考线，但这种方法存在两个主要问题。其一，人工点选效率较低、耗时较长；其二，参考线需要处于车道线的最中央，但人工点选生成的参考线是有一定宽度的，不可能精确的处于车道线的最中央，因此会造成误差。

本方案在得到道路点云数据后，通过人眼识别到属于车道线的点云数据，即通过人眼得到预确认的车道线，再从预确认的车道线上随意选取两个点作为起始点和根节点。后续将根节点作为下一个节点的搜索依据，通过计算机算法追踪搜索参考线上的其他节点。

从道路点云数据中点选整个车道线会出现误差，但从车道线上选取两个点是容易且准确的，相比于计算机选取，这种方式利用人的技术储备点选，提高了点选效率和准确性，为后续进行追踪搜索提供了准确的开端。

可以理解，也可以采用计算机获得预确认的车道线，也可以采用计算机从预确认的车道线中选取起始点和根节点，本方案对此不做具体限定。

具体地，假设起始点为P1，根节点为P2，向量的方向必须与采集车的行车方向一致，这样才能使后续追踪得到的参考线的方向与行车方向一致。

图2为本发明实施例提供的一种车道线示意图。

如图2所示，道路点云数据中包括三条车道线，最右侧的车道线被选中为预确认的车道线，该预确认的车道线将被作为待提取的参考线。从预确认的车道线上选取两个点，作为起始点和根节点。

步骤103、获取由所述采集车的行驶轨迹点构成的第一轨迹数据，并从所述第一轨迹数据中确定与所述根节点最近的第一轨迹点。

在追踪搜索参考线的下一个节点过程中，不清楚下一个节点的方向和位置，有可能道路已经转弯了，但路面点云数据还在延伸。

采集车在路面上行驶过程中，会形成一系列的GPS轨迹点。采集车的行驶轨迹与道路的前进方向是一致的，采集车行驶轨迹可以为搜索下一个节点提供指引。因此，可以采用行驶轨迹作为搜索下一个节点的重要参考，以提高搜索效率、防止搜索方向偏离。

本发明实施例中，可以每隔一定周期（如0.1秒）获取一次采集车的GPS轨迹点坐标，得到第一轨迹数据。第一轨迹数据具体包括各个轨迹点坐标和对应的采集时间戳。

路面点云数据具体包括路面上各个对象的三维坐标和采集时间戳，因此，可以确定根节点对应在路面维度的二维坐标。利用根节点的二维坐标和各个轨迹点的坐标，确定各个轨迹点与根节点的欧式距离，选取欧式距离最小的轨迹点作为第一轨迹点。

示例性地，假设选取的起始点和根节点分别为P1和P2，确定各个轨迹点与P2点的欧式距离，并从中选取与P2点的欧式距离最小的轨迹点G1，作为第一轨迹点。

步骤104、以所述第一轨迹点作为中心点，预加载所述中心点的预设半径范围内的所述路面点云数据，并对预加载的所述路面点云数据进行去噪处理，得到第一点云数据。

将第一轨迹点作为预加载的点云中心点，设定加载点云的半径R1，读取该半径范围内的路面点云数据。

由于路面点云数据中可能包括积水、落叶等噪声，因此，需要对读取的路面点云数据进行噪声滤除，得到第一点云数据。

步骤105、获取所述根节点的采集时间戳，并在所述采集时间戳的预设时间范围内有对应的所述第一轨迹数据的情况下，基于所述第一轨迹点的前进方向，确定新节点的搜索方向。

采集车的行驶轨迹与道路的前进方向是一致的，因此，某时刻的采集车的行驶轨迹与该时刻的车道线的前进方向是一致的，采集车行驶轨迹可以为搜索参考线的下一个节点提供指引。

首先从根节点的点云信息中获取采集时间戳，确认该采集时间戳的预设时间范围内是否存在第一轨迹数据。预设时间范围具体可以取0.1秒、0.05秒等较小的时间范围。

若预设时间范围内存在第一轨迹数据，则可以以第一轨迹数据的前进方向作为确定搜索方向的参考依据。

步骤106、从所述根节点开始，按照所述搜索方向和预设的搜索步长，从所述第一点云数据中搜索所述新节点。

在确认了搜索的方向后，需要设置搜索步长。搜索步长可以设置为一个固定值，也可以设置为一个随着搜索情况改变的值。此外，搜索方向也可以设置为随着搜索情况改变。

从根节点开始，根据搜索方向和搜索步长，从第一点云数据中搜索新节点。

步骤107、将所述第一轨迹点更新为所述新节点，以所述新节点为起始，继续执行所述去噪处理和搜索下一个新节点的操作。

在找到新节点之后，以新节点作为中心点，执行去噪处理和搜索下一个新节点的操作。

具体地，以新节点为中心点，预加载新节点的预设半径范围内的路面点云数据，并对预加载的路面点云数据进行去噪处理，得到第二点云数据；获取新节点的采集时间戳，并在采集时间戳的预设时间范围内有对应的第一轨迹数据的情况下，基于第一轨迹点的前进方向，确定下一个新节点的搜索方向；

从新节点开始，按照下一个新节点的搜索方向和预设的搜索步长，从第二第一点云数据中搜索下一个新节点。

不断重复前述步骤，直至触发停止条件，则停止搜索。在多次搜索过程中，得到依次向车道线前进方向延伸的多个新节点。

步骤108、基于所述起始点、所述根节点和多个搜索得到的新节点提取生成车道参考线。

将起始点、根节点以及搜索得到的各个新节点从道路点云数据中提取出来，并将各节点进行连接，得到车道参考线。

综上，在本发明实施例中，获取路面点云数据，所述路面点云数据是采集车在行驶过程中采集的路面上的点云数据，所述路面上包括多条车道线；从所述路面点云数据中获取起始点和根节点；获取由所述采集车的行驶轨迹点构成的第一轨迹数据，并从所述第一轨迹数据中确定与所述根节点最近的第一轨迹点；以所述第一轨迹点作为中心点，预加载所述中心点的预设半径范围内的所述路面点云数据，并对预加载的所述路面点云数据进行去噪处理，得到第一点云数据；获取所述根节点的采集时间戳，并在所述采集时间戳的预设时间范围内有对应的所述第一轨迹数据的情况下，基于所述第一轨迹点的前进方向，确定新节点的搜索方向；从所述根节点开始，按照所述搜索方向和预设的搜索步长，从所述第一点云数据中搜索所述新节点；将所述第一轨迹点更新为所述新节点，以所述新节点为起始，继续执行所述去噪处理和搜索下一个新节点的操作；基于所述起始点、所述根节点和多个搜索得到的新节点提取生成车道参考线。上述方法采用计算机算法进行新节点的搜索，提高了参考线的生成效率，从预确认的车道线对应的点云数据中选取参考线的起始点和根节点，提高了起始点和根节点的选取精度；并且，基于第一轨迹数据的前进方向，确定参考线的新节点的搜索方向，使得参考线的搜索方向与采集车的行驶方向保持一致，保证了搜索方向的准确性。

在一种可能的实施方式中，如图3所示，在步骤105中，基于所述第一轨迹点的前进方向，确定新节点的搜索方向，包括：

步骤1051、从所述第一轨迹数据中获取与所述第一轨迹点前后相邻的两个点，得到第一前点和第一后点，所述第一前点为所述第一轨迹点对应的上一时刻的轨迹点，所述第一后点为所述第一轨迹点对应的下一时刻的轨迹点。

第一轨迹数据包括各个轨迹点坐标及采集时间戳，根据采集时间戳可以得到第一轨迹点上一时刻的轨迹点坐标和下一时刻的轨迹点坐标，即第一前点的坐标和第一后点的坐标。

步骤1052、获取由所述起始点与所述根节点确定的第一方向、由所述第一前点与所述第一后点确定的第二方向之间的夹角。

具体地，由起始点P1和根节点P2确定的第一方向为。

设第一轨迹点G1的第一前点为G0，第一后点为G2，则由第一点前与第一后点确定的第二方向为。

与之间的夹角为第一方向与第二方向之间的夹角。

步骤1053、确定所述第一轨迹点与所述根节点之间的第一距离。

第一轨迹点G1和根节点P2之间的第一距离可以用欧式距离来计算。

步骤1054、基于所述第一距离、所述夹角确定新节点的搜索方向。

第一距离表明第一轨迹点与根节点之间的距离，夹角表明第一轨迹点的前进方向与起始点和根节点确定的方向之间的方向偏差，通过距离和夹角可以确定第一轨迹点属于车道线上的可能性。若第一轨迹点属于车道线的可能性较小，则放弃使用第一轨迹点来确定新节点的搜索方向，若第一轨迹点属于车道线的可能性较大，则可以利用第一轨迹点来确定新节点的搜索方向。

在步骤1051-步骤1054中，基于所述第一距离、所述夹角确定新节点的搜索方向，使得新节点的搜索方向可以根据第一轨迹点与根节点的距离决定，并同时根据第一方向与第二方向的夹角决定，使得搜索方向的决定更为科学、合理，保证了搜索方向确定的准确性。

在一种可能的实施方式中，步骤1054包括：

步骤10541、若所述第一距离大于或等于预设第一距离阈值，或所述夹角大于或等于预设夹角阈值，则将所述起始点和所述根节点确定的方向确定为新节点的搜索方向；所述第一距离阈值为P个轨迹数据采集周期内所述采集车行驶的距离。

在采集车行驶过程中，每隔一定周期（如0.1秒）获取一次采集车的GPS轨迹点坐标，得到第一轨迹数据。第一距离阈值为P个轨迹数据采集周期内所述采集车行驶的距离，若采集车在一个轨迹数据采集周期内行驶的距离平均为L米，则P个轨迹数据采集周期内采集车行驶的距离为P*L米，则第一距离阈值为P*L米。示例性地，第一距离阈值可以设置为6米。

若第一距离大于或等于预设第一距离阈值，则表明第一轨迹点与起始点之间的第一距离较大，这样，第一轨迹点属于起始点所在的车道线上的可能性较小，第一轨迹点的前进方向不能作为新节点的搜索方向。

类似地，若夹角大于或等于预设夹角阈值，则表明夹角过大，第一方向与第二方向之间的偏差较大，这样，第一轨迹点的前进方向与根节点和起始点所在的车道线的方向偏差较大，则第一轨迹点的前进方向不能作为新节点的搜索方向。

在出现上述两种情况时，将起始点和根节点确定的方向作为新节点的搜索方向。

示例性地，第一距离阈值可以设置为6米，夹角阈值可以设置为10度。那么，当第一轨迹点G1和根节点P2之间的第一距离大于或等于6米，或与之间的夹角大于或等于10度时，将方向确定为新节点的搜索方向。

步骤10542、若所述第一距离小于预设第一距离阈值，且所述夹角小于预设夹角阈值，则基于所述第一轨迹点、所述第一前点、所述第一后点确定所述新节点的搜索方向。

反之，若所述第一距离小于预设第一距离阈值，则表明第一轨迹点与起始点之间的第一距离较小；同时，若所述夹角小于预设夹角阈值，则表明第一轨迹点与起始点之间的夹角较小。这样，第一轨迹点的前进方向与根节点和起始点所在的车道线的方向较为一致，第一轨迹点属于起始点所在的车道线上的可能性较大，可基于第一轨迹点、第一前点、第一后点确定新节点的搜索方向。

在一种可能的实施方式中，步骤10542包括以下步骤：

步骤105421、确定所述第一后点与所述第一轨迹点之间的第二距离。

第一后点G2与第一轨迹点G1之间的第二距离可以采用欧式距离计算。

步骤105422、若所述第二距离大于或等于预设第二距离阈值，则将由所述第一前点与所述第一后点确定的方向作为所述新节点的搜索方向；所述第二距离阈值为Q个轨迹数据采集周期内所述采集车行驶的距离，Q＞P。

第二距离阈值为Q个轨迹数据采集周期内所述采集车行驶的距离，Q＞P，即第二距离阈值大于第一距离阈值。示例性地，第二距离阈值可以设置为10米。

当第二距离大于或等于第二距离阈值时，表明在采集车行驶过第一轨迹点之后，时间过了Q个采集周期，才获取到采集车的下一个GPS轨迹点，即第一后点。说明在中间的Q个轨迹数据采集周期内，采集设备未采集到轨迹点，轨迹数据在这Q个轨迹数据采集周期内不正常。发生这种情况的原因可能是采集设备被遮挡，或该路段无网络信号。

这时，第一后点可能也是有误差的，若以第一后点为标准来确定搜索方向，可能会使搜索方向产生较大的误差。所以，此种情况下以第一轨迹点作为标准来确定搜索方向，即用第一轨迹点对应的第一前点和第一后点确定的方向作为新节点的搜索方向。

步骤105423、若所述第二距离小于所述第二距离阈值，则将所述第一轨迹点更新为所述第一后点，并将由所述第一后点的前点和后点确定的方向作为所述新节点的搜索方向。

若第二距离小于第二距离阈值，表明在采集车行驶过第一轨迹点之后，在Q个轨迹数据采集周期内，采集设备已经采集到了第一后点，说明该路段的轨迹数据采集情况正常。在轨迹数据正常的情况下，第一后点的轨迹坐标也是正常的。此时，可以将第一轨迹点更新为第一后点，以第一后点为标准来确定新节点的搜索方向。

具体地，若第一后点的前点为G3，第一后点的后点为G4，则将方向作为新节点的搜索方向。

在一种可能的实施方式中，在步骤105中获取根节点的采集时间戳之后，还包括：

步骤109、在所述采集时间戳的预设时间范围内无对应的第一轨迹数据的情况下，将由所述起始点和所述根节点确定的方向，作为新节点的搜索方向。

某时刻的采集车的行驶轨迹与该时刻的车道线的前进方向是一致的，采集车行驶轨迹可以为搜索参考线的下一个节点提供指引。

但是，在采集时间戳的预设时间范围内无对应的采集车行驶轨迹数据的情况下，无法用采集车行驶轨迹作为搜索下一个节点的指引。

因此，这种情况下，仍然采用起始点和根节点确定的方向，作为新节点的搜索方向。

在一种可能的实施方式中，如图4所示，步骤104中对预加载的所述路面点云数据进行去噪处理，包括：

步骤1041、将预加载的所述路面点云数据按照点云反射强度进行排序。

路面点云数据中包括路面上的一切对象的点云信息，如普通路面、路面上的车道线、路面上的异物等。由于车道参考线是基于车道线的点云数据生成的，因此，将普通路面和路面上的异物对应的点云数据作为噪声进行滤除，仅保留车道线对应的点云数据，可以最大程度提高路面点云数据的纯度，加快后续新节点的搜索效率。

不同对象的点云反射强度不同，普通路面的反射强度一般小于车道线反射强度，积水、落叶、乘客遗落物等异物的反射强度一般大于车道线反射强度。因此，为了滤除噪声，可以先对路面点云数据按照点云反射强度进行排序。

点云反射强度，即为点云亮度，点云数据中一般本身包括点云的反射强度值。若未包括点云反射强度值，也可以根据点云数据中包括的点云的RGB值计算得到。

步骤1042、从所述路面点云数据中滤除所述排序在前N%的点云数据和所述排序在后M%的点云数据，得到初处理点云数据；所述排序在前N%的点云数据为路面上的异物对应的点云数据，所述排序在后M%的点云数据为普通路面对应的点云数据，所述初处理点云数据为车道线对应的点云数据。

可以根据仿真实验测定路面异物的反射强度均值、普通路面的反射强度均值和车道线的反射强度均值。

可以发现，相比于车道线，积水、落叶、乘客遗落物等路面异物的反射强度较高，但并不会比车道线的反射强度高太多；而普通路面的点云强度一般比车道线低的比较多。

根据实验确定M和N的具体取值，并取排序在前N%的点云数据，作为路面异物对应的点云数据，取排序在后M%的点云数据，作为普通路面对应的点云数据。示例性地，N取值为5，M取值为60。

步骤1043、将所述初处理点云数据所对应的空间区域按照预设网格尺寸进行划分，得到多个点云网格。

初处理点云数据将大部分的噪声进行了滤除，初处理点云数据对应的空间区域中，含点云数据较多的区域为车道线对应区域，含点云数据较少或不包含点云数据的区域为普通路面或异物等噪声对应的点云数据。

为了进一步把车道线点云数据的边缘补齐，进一步把噪声点云数据与车道线点云数据分开，可以对初处理点云数据划分网格，进行腐蚀膨胀处理。

可以将初处理点云数据对应的空间区域按照预设的网格尺寸划分为多个网格。预设网格尺寸可以根据点云密度和点云总数进行自行设定，例如可以设置为0.08米*0.08米。

步骤1044、根据所述点云网格中包括的点云数量与预设的第一数量阈值的大小关系，对各个所述点云网格进行初始二值化标记，并根据所述初始二值化标记结果对所述多个点云网格进行腐蚀膨胀处理，得到包括目标二值化标记结果的多个点云网格。

第一数量阈值根据网格尺寸进行预设，示例性地，在网格尺寸为0.08米*0.08米的情况下，第一数量阈值设置为1。

具体地，将点云数量小于第一数量阈值的点云网格和点云数量大于或等于第一数量阈值的点云网格，分别标记为不同的两个值，例如分别标记为0和1，得到初始二值化标记结果。

腐蚀的作用是消除物体边界点，使目标缩小，可以消除小于结构元素的噪声点；膨胀的作用是将与物体接触的所有背景点合并到物体中，使目标增大，可添补目标中的空洞。

根据初始二值化标记结果对多个点云网格进行腐蚀膨胀处理，目的是把黏连的、离散的点云网格区域化，把主体对象的边缘补齐，把不同对象分开。这样做完腐蚀膨胀后，保证车道线对应的点云网格中没有没有噪声对象黏连，保证车道线点云数据的纯度。

步骤1045、根据所述目标二值化标记结果，获取大于或等于所述第一数量阈值的目标点云网格，并将所述目标点云网格中的点云数据确定为第一点云数据。

做完腐蚀膨胀后，部分点云网格的二值化标记结果发生了变化。例如，有的由1变为0，有的由0变为1。若0表示点云数量小于第一数量阈值的点云网格，1表示点云数量大于或等于第一数量阈值的点云网格。则在目标二值化标记结果中，标记为1的点云网格，其基本可以确定是车道线对应的点云网格。

在步骤1041-步骤1045中，通过去噪处理提高了预加载的路面点云数据的数据纯度，得到的第一点云数据是车道线对应的点云数据，后续直接从第一点云数据中进行新节点搜索，有助于加快数据处理效率。

在一种可能的实施方式中，步骤1044包括：

步骤10431、将所述点云网格中包括的点云数量小于第一数量阈值的点云网格标记为P，并将所述点云数量大于或等于所述第一数量阈值的点云网格标记为Q，得到初始二值化标记结果。

若点云网格中包括的点云数量小于第一数量阈值th1，可以将该点云网格标记为P，P仅作为标识，无实际意义。若点云网格中包括的点云数量大于或等于th1，可以将该点云网格标记为Q，Q仅作为标识，无实际意义。

这样，根据对点云网格的标记，可以判定点云网格中的点云数量与第一数量阈值的关系。

步骤10432、根据所述初始二值化标记结果，对所述多个点云网格分别进行多次腐蚀处理和多次膨胀处理，得到包括目标二值化标记结果的多个点云网格。

点云网格的二值化标记结果是指标记有P或Q的多个网格。对标记有P或Q的多个网格进行多次腐蚀以及多次膨胀处理之后，可以进一步扩大一定范围内同一种标记结果较多的网格范围，并缩小同一种标记结果较少的网格范围。这样，若某个区域范围内P标记较多，则标记为P的区域范围会被膨胀，即扩大，若某个区域范围内P标记较少，则标记为P的区域范围会被腐蚀，即缩小。对于Q标记的范围，也是如此。

这样，就把黏连的点云数据和离散的点云数据进行了区域化，把车道线对象的边缘补齐，把噪声对象与车道线对象分开。这样做完腐蚀膨胀后，保证车道线对应的点云网格中没有没有噪声对象黏连，保证车道线点云数据的纯度。

步骤10433、将所述目标二值化标记结果为Q的点云网格中的点云数据确定为第一点云数据。

Q表示点云数量大于或等于第一数量阈值的点云网格。在目标二值化标记结果中，标记为Q的点云网格，其基本可以确定是车道线对应的点云网格。

将目标二值化标记结果为Q的点云网格中的点云数据确定为第一点云数据，用作接下来参考线生成中的节点追踪。

在一种可能的实施方式中，如图5所示，步骤106包括：

步骤1061、从所述根节点开始，按照所述搜索方向和预设的搜索步长，从所述第一点云数据中确定下一个节点。

在上一个步骤确定了搜索方向之后，按照预设的搜索步长，从根节点P2开始进行下一个节点的搜索，搜索方向设为，假设搜索到的下一个节点为P3，则搜索到的下一个节点满足以下公式：

（1）

步骤1062、确定所述下一个节点是否满足目标条件，所述目标条件为：所述节点所属的点云网格中的点云的数量大于或等于预设的第二数量阈值；所述第二数量阈值大于所述第一数量阈值。

在按照搜索方向和搜索步长，从第一点云数据中搜索到下一个节点之后，为了防止搜索到的下一个节点是噪声点云，可以利用目标条件进行验证。

由于大部分噪声点云在前面去噪处理的步骤已经被剔除了，第一点云数据中若存在少量噪声点云，这些少量噪声点云的位置附近点云数据量会非常少。因此可以设置第二数量阈值，根据下一个节点所属的点云网格中的点云数量与第二数量阈值的大小关系，判定下一个节点是否为噪声点云。

第二数量阈值大于第一数量阈值，示例性地，若第一数量阈值为1，则第二数量阈值可以设置为3。

在步骤104去噪处理的步骤中，采用较小的第一数量阈值来滤除噪声点云，并没有选用较大的第二数量阈值，是为了避免将车道线上的点云数据认为是噪声点云剔除掉。这样，采集较小的第一数量阈值导致第一点云数据中可能会混有少量的噪声点云，所以此处采用第二数量阈值判定下一个节点是否为噪声点云。采用这种分步骤加阈值的方法，使得方案的可行性、适用性更高。

步骤1063、若所述下一个节点满足所述目标条件，则确定所述点云网格中的各个点云的坐标值的平均值，并将所述平均值作为新节点的坐标值。

若下一个节点P3满足目标条件，则认为下一个节点所属的点云网格位于车道线上。

若下一个节点正好在车道线的边缘，则会导致这个点云网格中属于车道线的区域点云数量多，不属于车道线的区域点云数量少，而参考线最好位于车道线的最中间，若以点云网格中的某一个点云作为新节点，可能会导致参考线偏离车道线。为了避免出现这种参考线偏离的情况，取点云网格中各个点云坐标的平均值，作为新节点的坐标值。

在一种可能的实施方式中，在步骤1062之后，还包括：

步骤1064、若所述下一个节点不满足所述目标条件，则增加搜索步长，继续从所述第一点云数据中确定新的下一个节点；

步骤1065、确定所述新的下一个节点是否满足所述目标条件；

步骤1066、若不满足所述目标条件，则继续增加搜索步长，继续从所述第一点云数据中确定新的下一个节点，直到所述新的下一个节点满足所述目标条件，或本次搜索达到预设的停止条件。

在步骤1064-步骤1066中，若下一个节点不满足目标条件，则在原搜索方向上加上搜索步长ψ，重新计算下一个节点P4，并确定下一个节点P4是否满足目标条件。

若P4满足目标条件，则将下一个节点P4所属的点云网格的点云坐标平均值作为新节点的坐标。此后，获取所述该新节点的采集时间戳，并在所述采集时间戳的预设时间范围内有对应的所述第一轨迹数据的情况下，基于所述第一轨迹数据的前进方向，确定下一个新节点的搜索方向，即从步骤105开始重新生成搜索方向，进行后续搜索步骤。

若P4不满足目标条件，则继续加大搜索步长进行搜索，直到搜索得到满足目标条件的下一个节点。

若整个第一点云数据中均没有满足目标条件的点云网格，则停止本次搜索。这种情况下，对第一轨迹点T1进行更新，以第一轨迹点对应的下一时刻的轨迹点（即前述方案中的第一后点）作为中心点，预加载第一后点的预设半径范围内的路面点云数据，并从步骤104开始，重新执行去噪处理、确定搜索方向和节点搜索的步骤。

此外，在搜索过程中，若遇到预设的停止条件，则停止搜索。

在步骤1061-步骤1066中，上述实施例中，从根节点开始，按照搜索方向和预设的搜索步长确定目标节点，判断目标节点是否满足设定的目标条件，最终确定新节点的坐标值，在目标节点不满足设定的目标条件时，采用增大搜索步长的方法继续确定是否有满足条件的目标节点，在增大两次搜索步长后仍未发现有满足条件的目标节点，则对轨迹点T1进行更新，按照上述相同方法确定新节点的坐标值，提高了新节点的精确度。

在一种可能的实施方式中，所述预设的停止条件，包括：

所述新节点已到达路口；所述新节点的搜索方向与所述采集车的行车方向相反；所述新节点与最近的已有参考线起始节点之间的第三距离小于第三距离阈值；所述已有参考线为同向车道的参考线；所述新节点的搜索步长超过预设的步长阈值。

在本发明实施例中，若出现上述预设的停止条件中的任何一种条件，则停止新节点的搜索。

其中，若新节点已到达路口，说明该条车道线已到达端点，则可以停止对该车道线上的新节点的搜索。

若新节点的搜索方向与采集车的行车方向相反，即本次参考线选择节点确定的方向与采集车的行车轨迹相反，则说明新节点的搜索方向可能是错误的，需要重新设置搜索方向。

若新节点与最近的同向车道的已有参考线起始节点之间的第三距离小于第三距离阈值，则可以将当前搜索得到的参考线并入上述已有参考线中，无需再进行新节点的搜索。其中，第三距离阈值可以为设置为当前车道宽度的一半，例如当前车道宽度为4米，则第三距离阈值设置为2米。

若新节点与同向车道的参考线起始节点的距离小于当前车道宽度的一半，说明新节点是属于同向车道上的节点，那么当前搜索得到的参考线与同向车道是属于同一条参考线，可以将当前搜索得到的参考线并入同向车道的参考线中。

具体的，同向车道参考线的确定方法为：

将新节点更新为根节点，确定根节点附近是否存在已绘制的参考线，在存在已绘制参考线的情况下，将已有参考线的初始节点设为P',若存在与的夹角大于或等于90°，判断该已有参考线为对向车道的参考线，忽略该已有参考线。若与的夹角小于90°，则判断该已有参考线为同向车道的参考线。

若与的夹角小于90°，获取P2与最近的已有参考线起始节点之间的第三距离，可以将第三距离阈值设置为2米，若第三距离小于2米，将最后一次确定的新节点设为已有参考线的初始节点并停止新节点的搜索。

此外，若新节点的搜索步长超过预设的步长阈值，则说明搜索步长已经过长，误差太大，此时需要停止搜索。

在本实施例中，可以将步长阈值设置为4米，在搜索新节点的过程中，若新节点的搜索步长超过预设的步长阈值，则停止节点搜索。

在一种可能的实施方式中，在步骤108之后，还包括：

步骤110、基于所述车道参考线与各个所述车道线的位置关系、预设的车道宽度生成所述路面的多条车道线。

车道参考线为多条车道线中的其中一条，各条车道线之间是平行的位置关系，根据该车道参考线的位置，确定其他车道线与其之间的横向距离。然后采用对车道参考线复制平移的方式，可以生成其他的车道线。

图6为本发明实施例提供的一种车道参考线的提取装置的结构示意图。如图6所示，该车道参考线的提取装置200包括：

路面点云数据获取模块201，用于获取路面点云数据，所述路面点云数据是采集车在行驶过程中采集的路面上的点云数据，所述路面上包括多条车道线；

起始点根节点获取模块202，用于从所述路面点云数据中获取起始点和根节点，所述起始点和所述根节点为人工预确认的车道线上的两个点；所述起始点到所述根节点的方向与所述采集车的行车方向一致；所述人工预确认的车道线为所述多条车道线中的其中一条；

第一轨迹点确定模块203，用于获取由所述采集车的行驶轨迹点构成的第一轨迹数据，并从所述第一轨迹数据中确定与所述根节点最近的第一轨迹点；

去噪处理模块204，用于以所述第一轨迹点作为中心点，预加载所述中心点的预设半径范围内的所述路面点云数据，并对预加载的所述路面点云数据进行去噪处理，得到第一点云数据；

第一搜索方向确定模块205，用于获取所述根节点的采集时间戳，并在所述采集时间戳的预设时间范围内有对应的所述第一轨迹数据的情况下，基于所述第一轨迹点的前进方向，确定新节点的搜索方向；

新节点搜索模块206，用于从所述根节点开始，按照所述搜索方向和预设的搜索步长，从所述第一点云数据中搜索所述新节点；

更新模块207，用于将所述第一轨迹点更新为所述新节点，以所述新节点为起始，继续执行所述去噪处理和搜索下一个新节点的操作；

车道参考线生成模块208，用于基于所述起始点、所述根节点和多个搜索得到的新节点提取生成车道参考线。

可选的，所述第一搜索方向确定模块205具体用于：

从所述第一轨迹数据中获取与所述第一轨迹点前后相邻的两个点，得到第一前点和第一后点，所述第一前点为所述第一轨迹点对应的上一时刻的轨迹点，所述第一后点为所述第一轨迹点对应的下一时刻的轨迹点；

获取由所述起始点与所述根节点确定的第一方向、由所述第一前点与所述第一后点确定的第二方向之间的夹角；

确定所述第一轨迹点与所述根节点之间的第一距离；

基于所述第一距离、所述夹角确定新节点的搜索方向。

可选的，所述第一搜索方向确定模块205进一步用于：

若所述第一距离大于或等于预设第一距离阈值，或所述夹角大于或等于预设夹角阈值，则将所述起始点和所述根节点确定的方向确定为新节点的搜索方向；所述第一距离阈值为P个轨迹数据采集周期内所述采集车行驶的距离；

若所述第一距离小于预设第一距离阈值，且所述夹角小于预设夹角阈值，则基于所述第一轨迹点、所述第一前点、所述第一后点确定所述新节点的搜索方向。

可选的，所述第一搜索方向确定模块205进一步用于：

确定所述第一后点与所述第一轨迹点之间的第二距离；

若所述第二距离大于或等于预设第二距离阈值，则将由所述第一前点与所述第一后点确定的方向作为所述新节点的搜索方向；所述第二距离阈值为Q个轨迹数据采集周期内所述采集车行驶的距离，Q＞P；

若所述第二距离小于所述第二距离阈值，则将所述第一轨迹点更新为所述第一后点，并将由所述第一后点的前点和后点确定的方向作为所述新节点的搜索方向。

可选的，所述车道参考线的提取装置200还包括：

第二搜索方向确定模块，用于在所述采集时间戳的预设时间范围内无对应的第一轨迹数据的情况下，将由所述起始点和所述根节点确定的方向，作为新节点的搜索方向。

可选的，所述去噪处理模块204具体用于：

将预加载的所述路面点云数据按照点云反射强度进行排序；

从所述路面点云数据中滤除所述排序在前N%的点云数据和所述排序在后M%的点云数据，得到初处理点云数据；所述排序在前N%的点云数据为路面上的异物对应的点云数据，所述排序在后M%的点云数据为普通路面对应的点云数据，所述初处理点云数据为车道线对应的点云数据；

将所述初处理点云数据所对应的空间区域按照预设网格尺寸进行划分，得到多个点云网格；

根据所述点云网格中包括的点云数量与预设的第一数量阈值的大小关系，对各个所述点云网格进行初始二值化标记，并根据所述初始二值化标记结果对所述多个点云网格进行腐蚀膨胀处理，得到包括目标二值化标记结果的多个点云网格；

根据所述目标二值化标记结果，获取大于或等于所述第一数量阈值的目标点云网格，并将所述目标点云网格中的点云数据确定为第一点云数据。

可选的，所述去噪处理模块204进一步用于：

将所述点云网格中包括的点云数量小于第一数量阈值的点云网格标记为P，并将所述点云数量大于或等于所述第一数量阈值的点云网格标记为Q，得到初始二值化标记结果；

根据所述初始二值化标记结果，对所述多个点云网格分别进行多次腐蚀处理和多次膨胀处理，得到包括目标二值化标记结果的多个点云网格；

将所述目标二值化标记结果为Q的点云网格中的点云数据确定为第一点云数据。

可选的，所述新节点搜索模块206具体用于：

从所述根节点开始，按照所述搜索方向和预设的搜索步长，从所述第一点云数据中确定下一个节点；

确定所述下一个节点是否满足目标条件，所述目标条件为：所述节点所属的点云网格中的点云的数量大于或等于预设的第二数量阈值；所述第二数量阈值大于所述第一数量阈值；

若所述下一个节点满足所述目标条件，则确定所述点云网格中的各个点云的坐标值的平均值，并将所述平均值作为新节点的坐标值。

可选的，所述新节点搜索模块206具体用于：

若所述下一个节点不满足所述目标条件，则增加搜索步长，继续从所述第一点云数据中确定新的下一个节点；

确定所述新的下一个节点是否满足所述目标条件；

若不满足所述目标条件，则继续增加搜索步长，继续从所述第一点云数据中确定新的下一个节点，直到所述新的下一个节点满足所述目标条件，或本次搜索达到预设的停止条件。

可选的，所述车道参考线的提取装置200还包括：

车道线生成模块，用于基于所述参考线与各个所述车道线的位置关系、预设的车道宽度生成所述路面的多条车道线。

所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的车道参考线的提取装置的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。

在本发明提供的又一实施例中，还提供了一种电子设备，所述电子设备包括处理器和存储器，所述存储器中存储有至少一条指令或者至少一段程序，所述至少一条指令或者至少一段程序由所述处理器加载并执行以实现本发明实施例中所述的车道参考线的提取方法。

在本发明提供的又一实施例中，还提供了一种计算机可读存储介质，所述存储介质中存储有至少一条指令或者至少一段程序，所述至少一条指令或者至少一段程序由处理器加载并执行以实现本发明实施例中所述的车道参考线的提取方法。

上述主要从设备的角度对本发明实施例提供的方案进行了介绍。可以理解的是，设备为了实现上述功能，其包含了执行各个功能相应的硬件结构和/或软件模块。本领域技术人员应该很容易意识到，结合本文中所公开的实施例描述的各示例的算法步骤，本发明能够以硬件或硬件和计算机软件的结合形式来实现。某个功能究竟以硬件还是计算机软件驱动硬件的方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本发明的范围。

以上所述，仅为本发明的具体实施方式，但本发明的保护范围并不局限于此，任何在本发明揭露的技术范围内的变化或替换，都应涵盖在本发明的保护范围之内。因此，本发明的保护范围应以所述权利要求的保护范围为准。

Specific embodiment

The following will be combined with the accompanying drawings in the embodiments of the present invention, the technical solution in the embodiments of the present invention is clearly and completely described, obviously, the described embodiments are only a part of the embodiments of the present invention, not all embodiments. Based on embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative labor, fall within the scope of protection of the present invention.

Hereinafter, the terms "first" and "second" are for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, the features limited to "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present public embodiment, unless otherwise stated, "plurality" means two or more. Further, the use of "based on" or "based on" means open and inclusive, because processes, steps, calculations or other actions "based on" or "according to" one or more of the said conditions or values may, in practice, be based on additional conditions or exceed the stated values.

FIG 1 is one of the flowcharts of a lane reference line extraction method provided by an embodiment of the present invention, the method may comprise the following steps:

Step 101, obtain the pavement point cloud data, the pavement point cloud data is the point cloud data on the road surface collected by the collecting vehicle during driving, the pavement includes a plurality of lane lines.

The high-precision map collection vehicle collects the point cloud data of the entire road surface during the driving process on the road surface to obtain the road surface point cloud data. The pavement point cloud data includes the point cloud information of all objects on the pavement, such as the pavement, lane lines on the pavement, foreign objects on the pavement, and so on.

Each point in the pavement point cloud data includes information such as three-dimensional coordinates, acquisition timestamps, color information (RGB values), and reflection intensity information (IQteQsity).

Step 102, obtain the starting point and the root node from the pavement point cloud data, the starting point and the root node are two points on the manually preconfirmed lane line; The direction from the starting point to the root node is consistent with the direction of travel of the collection vehicle; The manually preconfirmed lane line is one of the plurality of lane lines.

The start point and root node are used as the first two points to generate a lane reference line, hereinafter referred to as a guide. The start point and root nodes here are extracted from the manually preconfirmed lane lines, so the reference line that this scenario ultimately generates is one of the lane lines.

The guides in this scenario are reference identifiers that generate other lane lines on the pavement and are used as direction and length references for generating other lane lines. After obtaining the reference lines, other lane lines on the road surface can be obtained by copying the panning method.

In related technology, a lane line is obtained by all manual clicks, and the lane line is used as a reference line to generate other lane lines, but there are two main problems in this method. First, manual selection is less efficient and time-consuming; Second, the reference line needs to be in the very center of the lane line, but the reference line generated by manual clicking has a certain width, and it is impossible to be accurately in the center of the lane line, so it will cause errors.

In this scheme, after obtaining the road point cloud data, the point cloud data belonging to the lane line is identified by the human eye, that is, the preconfirmed lane line is obtained through the human eye, and then two points are randomly selected from the preconfirmed lane line as the starting point and root node. Subsequently, the root node is used as the search basis for the next node, and other nodes on the search reference line are tracked by computer algorithms.

Selecting the entire lane line from the road point cloud data will cause errors, but it is easy and accurate to select two points from the lane line, compared with computer selection, this method uses human technical reserve point selection, improves the efficiency and accuracy of point selection, and provides an accurate starting point for subsequent tracking search.

It is understood that a computer can also be used to obtain preconfirmed lane lines, or a computer can be used to select the starting point and root node from the preconfirmed lane lines, and this scheme does not specifically limit this.

Specifically, assuming that the starting point is P1 and the root node is P2, the direction of the vector must be consistent with the driving direction of the acquisition vehicle, so that the direction of the reference line obtained by subsequent tracking can be consistent with the driving direction.

FIG 2 is a schematic view of a lane line provided by an embodiment of the present invention.

As shown in Figure 2, the road point cloud data includes three lane lines, and the rightmost lane line is selected as the preconfirmed lane line, which will be used as the reference line to be extracted. Pick two points from the preconfirmed lane line to serve as the start point and root node.

Step 103, obtain the first trajectory data composed of the driving trajectory point of the acquisition vehicle, and determine the first trajectory point closest to the root node from the first trajectory data.

In the process of tracking the next node of the search reference line, the direction and location of the next node is not clear, it is possible that the road has turned, but the pavement point cloud data is still extended.

During the process of the collection vehicle driving on the road, a series of GPS trajectory points will be formed. The trajectory of the collection vehicle is consistent with the direction of the road, and the driving trajectory of the collection vehicle can provide guidance for the search for the next node. Therefore, the driving track can be used as an important reference for searching for the next node to improve the search efficiency and prevent the search direction from deviating.

In an embodiment of the present invention, the GPS trajectory point coordinates of the acquisition vehicle can be obtained every certain period (such as 0.1 seconds) to obtain the first trajectory data. The first trajectory data specifically includes the coordinates of each trajectory point and the corresponding acquisition timestamp.

The pavement point cloud data specifically includes the three-dimensional coordinates and acquisition timestamps of each object on the pavement, so the two-dimensional coordinates corresponding to the root node in the pavement dimension can be determined. The two-dimensional coordinates of the root node and the coordinates of each trajectory point are used to determine the Euclidean distance between each trajectory point and the root node, and the trajectory point with the smallest Euclidean distance is selected as the first trajectory point.

For example, suppose that the starting point and root node are selected P1 and P2, respectively, determine the Euclidean distance between each trajectory point and P2 point, and select the trajectory point G1, which has the smallest Euclidean distance from P2 point, as the first trajectory point.

Step 104, taking the first trajectory point as the center point, preload the pavement point cloud data within the preset radius of the center point, and denoise the preloaded pavement point cloud data to obtain the first point cloud data.

Take the first trajectory point as the center point of the preloaded point cloud, set the radius R1 of the loaded point cloud, and read the pavement point cloud data within this radius.

Since the pavement point cloud data may include noise such as standing water and falling leaves, it is necessary to filter out the noise of the read pavement point cloud data to obtain the first point cloud data.

Step 105, obtain the acquisition timestamp of the root node, and in the case of the corresponding first trajectory data within the preset time range of the acquisition timestamp, based on the forward direction of the first trajectory point, determine the search direction of the new node.

The driving trajectory of the collection vehicle is consistent with the forward direction of the road, so the driving trajectory of the collection vehicle at a certain time is consistent with the forward direction of the lane line at that moment, and the driving trajectory of the collection vehicle can provide guidance for the next node of the search reference line.

Firstly, the acquisition timestamp is obtained from the point cloud information of the root node to confirm whether the first trajectory data exists within the preset time range of the acquisition timestamp. The preset time range can be taken from a smaller time range such as 0.1 seconds and 0.05 seconds.

If the first trajectory data exists within the preset time range, the forward direction of the first trajectory data can be used as the reference basis for determining the search direction.

Step 106, starting from the root node, according to the search direction and the preset search step, search for the new node from the first point cloud data.

After confirming the direction of the search, you need to set the search step size. The search step size can be set to a fixed value or to a value that changes depending on the search situation. In addition, the search direction can also be set to change with the search situation.

Starting from the root node, search for new nodes from the first point cloud data based on search direction and search step.

Step 107, the first trajectory point is updated to the new node, starting with the new node, continue to perform the denoising processing and search for the next new node operation.

After the new node is found, denoising and searching for the next new node are performed with the new node as the central point.

Specifically, with the new node as the central point, the pavement point cloud data within the preset radius of the new node is preloaded, and the preloaded pavement point cloud data is denoising to obtain the second point cloud data. Obtain the acquisition timestamp of the new node, and determine the search direction of the next new node based on the forward direction of the first trajectory point when the corresponding first trajectory data is available within the preset time range of the acquisition timestamp.

Starting from the new node, search for the next new node from the second first point cloud data according to the search direction of the next new node and the preset search step.

Repeat the preceding steps until the stop condition is triggered, and then stop the search. During multiple searches, multiple new nodes are obtained that extend sequentially in the direction of the lane line.

Step 108, based on the starting point, the root node and a plurality of search new nodes extracted to generate a lane reference line.

The start point, root node and each new node obtained by the search are extracted from the road point cloud data, and the nodes are connected to obtain the lane guide line.

In summary, in an embodiment of the present invention, the pavement point cloud data is obtained, the pavement point cloud data is the point cloud data on the road surface collected by the collecting vehicle during driving, and the pavement comprises a plurality of lane lines; Obtain the starting point and root node from the pavement point cloud data; Obtain the first trajectory data composed of the driving trajectory point of the acquisition vehicle, and determine the first trajectory point closest to the root node from the first trajectory data; Taking the first trajectory point as the center point, the pavement point cloud data within the preset radius of the center point is preloaded, and the pre-loaded pavement point cloud data is denoising to obtain the first point cloud data; Obtain the acquisition timestamp of the root node, and in the case of the corresponding first trajectory data within the preset time range of the acquisition timestamp, the search direction of the new node is determined based on the forward direction of the first trajectory point; Starting from the root node, the new node is searched from the first point cloud data according to the search direction and the preset search step; update the first trajectory point to the new node, starting with the new node, continue to perform the denoising processing and search for the next new node; Route reference lines are generated based on the origin point, the root node and a plurality of new nodes obtained by searching. The above method uses computer algorithm to search for new nodes, which improves the generation efficiency of reference lines, selects the starting point and root node of the reference line from the point cloud data corresponding to the preconfirmed lane line, and improves the selection accuracy of the starting point and root node. Moreover, based on the forward direction of the first trajectory data, the search direction of the new node of the reference line is determined, so that the search direction of the reference line is consistent with the driving direction of the collection vehicle, which ensures the accuracy of the search direction.

In one possible embodiment, as shown in FIG. 3, in step 105, based on the forward direction of the first trajectory point, the search direction of the new node is determined, comprising:

Step 1051, from the first trajectory data to obtain two points adjacent to the first trajectory point before and after, obtain the first front point and the first back point, the first front point is the trajectory point of the previous moment corresponding to the first trajectory point, the first rear point is the trajectory point of the next moment corresponding to the first trajectory point.

The first trajectory data includes the coordinates of each trajectory point and the acquisition timestamp, according to the acquisition timestamp, the locus point coordinates of the previous moment of the first trajectory point and the coordinates of the trajectory point of the next moment can be obtained, that is, the coordinates of the first front point and the coordinates of the first back point.

Step 1052, obtaining the angle between the first direction determined by the starting point and the root node, and the second direction determined by the first pre-point and the first post-point.

Specifically, the first direction determined by the start point P1 and the root node P2 is.

Let the first anterior point of the first trajectory point G1 be G0 and the first posterior point be G2, then the second direction determined by the first precedent and first posterior points is.

The angle between and is the angle between the first and second directions.

Step 1053, determine the first distance between the first trajectory point and the root node.

The first distance between the first trajectory point G1 and the root node P2 can be calculated using the Euclidean distance.

Step 1054, based on the first distance, the angle determines the search direction of the new node.

The first distance indicates the distance between the first trajectory point and the root node, and the angle indicates the direction deviation between the forward direction of the first trajectory point and the direction determined by the starting point and the root node, and the possibility that the first trajectory point belongs to the lane line can be determined by the distance and angle. If the first track point is less likely to belong to the lane line, abandon using the first track point to determine the search direction of the new node, and if the first track point is more likely to belong to the lane line, the first track point can be used to determine the search direction of the new node.

In steps 1051-step 1054, the search direction of the new node is determined based on the first distance and the angle, so that the search direction of the new node can be determined according to the distance between the first trajectory point and the root node, and at the same time according to the angle between the first direction and the second direction, so that the decision of the search direction is more scientific and reasonable, and the accuracy of the search direction determination is guaranteed.

In one possible embodiment, step 1054 comprises:

Step 10541, if the first distance is greater than or equal to the preset first distance threshold, or the angle is greater than or equal to the preset angle threshold, the starting point and the direction determined by the root node are determined as the search direction of the new node; The first distance threshold is the distance traveled by the acquisition vehicle in P trajectory data acquisition cycles.

During the driving process of the acquisition vehicle, the GPS trajectory point coordinates of the acquisition vehicle are obtained every certain cycle (such as 0.1 seconds) to obtain the first trajectory data. The first distance threshold is the distance traveled by the acquisition vehicle in P trajectory data collection cycles, if the average distance traveled by the acquisition vehicle in one trajectory data acquisition cycle is L meters, then the distance traveled by the acquisition vehicle in P trajectory data acquisition cycles is P*L meters, then the first distance threshold is P*L meters. For example, the first distance threshold can be set to 6 meters.

If the first distance is greater than or equal to the preset first distance threshold, it indicates that the first distance between the first track point and the starting point is large, so that the first track point is less likely to belong to the lane line where the start point is located, and the forward direction of the first track point cannot be used as the search direction of the new node.

Similarly, if the angle is greater than or equal to the preset angle threshold, it indicates that the angle is too large and the deviation between the first direction and the second direction is large, so that the forward direction of the first trajectory point deviates greatly from the direction of the lane line where the root node and the starting point are located, and the forward direction of the first trajectory point cannot be used as the search direction of the new node.

In both cases, the direction determined by the origin and root nodes is used as the search direction for the new node.

For example, the first distance threshold may be set to 6 meters, and the angle threshold may be set to 10 degrees. Then, when the first distance between the first trajectory point G1 and the root node P2 is greater than or equal to 6 meters, or the angle between the > and / is greater than or equal to 10 degrees, the direction is determined as the search direction of the new node.

Step 10542, if the first distance is less than the preset first distance threshold, and the angle is less than the preset angle threshold, the search direction of the new node is determined based on the first trajectory point, the first front point, and the first back point.

Conversely, if the first distance is less than the preset first distance threshold, it indicates that the first distance between the first trajectory point and the starting point is small; At the same time, if the angle is less than the preset angle threshold, it indicates that the angle between the first trajectory point and the starting point is small. In this way, the forward direction of the first trajectory point is more consistent with the direction of the lane line where the root node and the starting point are located, and the first trajectory point is more likely to belong to the lane line where the starting point is located, and the search direction of the new node can be determined based on the first trajectory point, the first front point, and the first back point.

In one possible embodiment, step 10542 comprises the following steps:

Step 105421, determine the second distance between the first post-point and the first trajectory point.

The second distance between the first posterior point G2 and the first trajectory point G1 can be calculated using the Euclidean distance.

Step 105422, if the second distance is greater than or equal to the preset second distance threshold, the direction determined by the first pre-point and the first post-point as the search direction of the new node; The second distance threshold is the distance traveled by the acquisition vehicle in Q trajectory data acquisition cycles, Q>P.

The second distance threshold is the distance traveled by the acquisition vehicle in Q trajectory data acquisition cycles, Q>P, that is, the second distance threshold is greater than the first distance threshold. For example, the second distance threshold can be set to 10 meters.

When the second distance is greater than or equal to the second distance threshold, it indicates that after the acquisition vehicle has driven through the first trajectory point, the time has passed Q collection cycles before the next GPS track point of the acquisition vehicle, that is, the first rear point. It shows that the acquisition device does not collect the trajectory point during the Q trajectory data collection cycle in the middle, and the trajectory data is abnormal in this Q trajectory data acquisition cycle. This can happen because the acquisition device is occluded, or there is no network signal on the road.

At this time, the first post-point may also have errors, and if the first post-point is used as the standard to determine the search direction, it may cause a large error in the search direction. Therefore, in this case, the first trajectory point is used as the standard to determine the search direction, that is, the direction determined by the first pre-point and the first post-point corresponding to the first trajectory point is used as the search direction of the new node.

Step 105423, if the second distance is less than the second distance threshold, the first trajectory point is updated to the first postpoint, and the direction determined by the front and back points of the first post-point is taken as the search direction of the new node.

If the second distance is less than the second distance threshold, it indicates that after the acquisition vehicle has driven through the first trajectory point, the acquisition device has collected the first rear point within Q trajectory data collection cycles, indicating that the trajectory data collection of this road section is normal. In the case of normal trajectory data, the trajectory coordinates of the first postpoint are also normal. At this point, you can update the first track point to the first postpoint, and use the first postpoint as the standard to determine the search direction of the new node.

Specifically, if the front point of the first postpoint is G3 and the postpoint of the first postpoint is G4, the direction is used as the search direction of the new node.

In one possible embodiment, after obtaining the acquisition timestamp of the root node in step 105, further comprises:

Step 109, in the case of no corresponding first trajectory data within the preset time range of the acquisition timestamp, the direction determined by the starting point and the root node is used as the search direction of the new node.

The trajectory of the collection vehicle at a certain moment is consistent with the forward direction of the lane line at that moment, and the driving trajectory of the collection vehicle can provide guidance for the next node of the search reference line.

However, if there is no corresponding vehicle trajectory data within the preset time range of the collection timestamp, the vehicle trajectory cannot be used as a guide to search for the next node.

Therefore, in this case, the direction determined by the origin and root nodes is still used as the search direction for the new node.

In one possible embodiment, as shown in FIG. 4, step 104 is a preloaded pavement point cloud data denoising process, comprising:

Step 1041, the preloaded pavement point cloud data is sorted according to the point cloud reflection intensity.

Pavement point cloud data includes the point cloud information of all objects on the pavement, such as ordinary pavement, lane lines on pavement, foreign objects on pavement, etc. Since the lane reference line is generated based on the point cloud data of the lane line, the point cloud data corresponding to the ordinary road surface and foreign objects on the road surface is filtered out as noise, and only the point cloud data corresponding to the lane line is retained, which can maximize the purity of the road surface point cloud data and accelerate the search efficiency of subsequent new nodes.

The reflection intensity of point clouds of different objects is different, the reflection intensity of ordinary pavement is generally less than that of lane lines, and the reflection intensity of foreign objects such as stagnant water, fallen leaves, and passenger leaves is generally greater than that of lane line reflection. Therefore, in order to filter out the noise, you can first sort the pavement point cloud data according to the point cloud reflection intensity.

Point cloud reflection intensity, that is, point cloud brightness, point cloud data generally includes the point cloud reflection intensity value. If the point cloud reflection intensity value is not included, it can also be calculated from the RGB value of the point cloud included in the point cloud data.

Step 1042, from the pavement point cloud data sorted in the first N% and the point cloud data sorted in the last M%, the initial processing point cloud data is obtained; The point cloud data sorted in the first N% is the point cloud data corresponding to the foreign object on the road surface, the point cloud data sorted in the last M% is the point cloud data corresponding to the ordinary road surface, and the initial processing point cloud data is the point cloud data corresponding to the lane line.

According to simulation experiments, the average reflection intensity of foreign objects on the pavement, the average reflection intensity of ordinary pavement and the average reflection intensity of lane lines can be determined.

It can be found that compared with lane lines, the reflection intensity of road surface foreign objects such as stagnant water, fallen leaves, and passenger remains is higher, but it is not much higher than that of lane lines; The point cloud intensity of ordinary pavement is generally lower than that of lane lines.

According to the experiment, the specific values of M and N were determined, and the point cloud data sorted in the top N% was taken as the point cloud data corresponding to the pavement foreign object, and the point cloud data sorted in the last M% was taken as the point cloud data corresponding to the ordinary pavement. For example, N takes 5 and M takes 60.

Step 1043, the spatial area corresponding to the initial processing point cloud data is divided according to the preset grid size to obtain a plurality of point cloud meshes.

In the spatial area corresponding to the primary processing point cloud data, the area with more point cloud data is the corresponding area of the lane line, and the area containing less point cloud data or not containing point cloud data is the point cloud data corresponding to noise such as ordinary road surface or foreign matter.

In order to further complete the edge of the lane line point cloud data and further separate the noise point cloud data from the lane line point cloud data, the initial processing point cloud data can be divided into grids and corroded and expanded.

The spatial area corresponding to the primary processing point cloud data can be divided into multiple meshes according to the preset grid size. The preset grid size can be set according to the point cloud density and the total number of point clouds, for example, it can be set to 0.08 m * 0.08 m.

Step 1044, according to the size relationship between the number of point clouds included in the point cloud grid and the preset first quantity threshold, each point cloud grid is initially binary marked, and the plurality of point cloud grids are corroded and expanded according to the initial binary labeling results, and a plurality of point cloud grids including target binary labeling results are obtained.

The first quantity threshold is preset according to the grid size, for example, when the grid size is 0.08 m * 0.08 m, the first quantity threshold is set to 1.

Specifically, the point cloud grid with the number of point clouds less than the first quantity threshold and the point cloud grid with the number of point clouds greater than or equal to the first number threshold are labeled with different two values, such as 0 and 1, respectively, to obtain the initial binary labeling results.

The effect of corrosion is to eliminate the boundary point of the object, so that the target is reduced, and the noise point that is smaller than the structural element can be eliminated; The effect of expansion is to merge all the background points in contact with the object into the object, making the target larger and filling the void in the target.

According to the initial binarization labeling results, the multiple point cloud grids are corroded and expanded, with the aim of regionalizing the adhesional and discrete point cloud meshes, completing the edges of the main objects and separating the different objects. After this is done for corrosion expansion, ensure that there is no noisy object adhesion in the point cloud grid corresponding to the lane line, and ensure the purity of the lane line point cloud data.

Step 1045, according to the target binarization labeling result, obtain a target point cloud grid greater than or equal to the first quantity threshold, and determine the point cloud data in the target point cloud grid as the first point cloud data.

After the corrosion expansion, the binarized marking results of some point cloud meshes change. For example, some change from 1 to 0, and some change from 0 to 1. If 0 indicates that the number of point clouds is less than the first number of threshold point cloud meshes, 1 means that the number of point clouds is greater than or equal to the first number of threshold point cloud meshes. In the target binary labeling results, the point cloud grid marked as 1 can basically be determined to be the point cloud mesh corresponding to the lane line.

In steps 1041-1045, the data purity of the preloaded pavement point cloud data is improved by denoising processing, and the first point cloud data obtained is the point cloud data corresponding to the lane line, and the subsequent search for new nodes directly from the first point cloud data helps to accelerate the data processing efficiency.

In one possible embodiment, step 1044 comprises:

Step 10431, the point cloud grid included in the point cloud grid is less than the first number threshold point cloud grid marked P, and the point cloud grid greater than or equal to the first number threshold point cloud grid is marked Q, the initial binary labeling result is obtained.

If the number of point clouds included in the point cloud grid is less than the first number threshold th1, the point cloud grid can be marked as P, which is only used as an identification and has no practical meaning. If the number of point clouds included in the point cloud grid is greater than or equal to th1, the point cloud grid can be marked as Q, and Q is only used as an identification and has no practical meaning.

In this way, based on the labeling of the point cloud mesh, the number of point clouds in the point cloud grid can be determined in relation to the first number threshold.

Step 10432, according to the initial binarization labeling results, the plurality of point cloud grids are subjected to multiple corrosion treatment and multiple expansion treatments, respectively, and a plurality of point cloud meshes including the target binarization labeling results are obtained.

The binary labeling result of a point cloud mesh refers to multiple meshes labeled P or Q. After multiple corrosion and multiple expansion treatments of multiple meshes marked with P or Q, the range of meshes with more results of the same marking can be further expanded within a certain range, and the range of meshes with fewer results of the same type of marking can be further expanded. In this way, if there are more P marks in a certain area, the area range marked P will be expanded, that is, expanded, and if there are fewer P marks in a certain area, the area range marked P will be corroded, that is, reduced. The same is true for the range of Q markers.

In this way, the adhesion point cloud data and discrete point cloud data are regionalized, the edges of the lane line object are completed, and the noise object is separated from the lane line object. After this is done for corrosion expansion, ensure that there is no noisy object adhesion in the point cloud grid corresponding to the lane line, and ensure the purity of the lane line point cloud data.

Step 10433, the point cloud data in the point cloud grid with the target binarization mark result of Q is determined to be the first point cloud data.

Q represents a point cloud mesh where the number of point clouds is greater than or equal to the first number threshold. In the target binary labeling result, the point cloud grid marked as Q can basically be determined to be the point cloud grid corresponding to the lane line.

The point cloud data in the point cloud grid with the target binarization mark result of Q is determined as the first point cloud data, which is used as the node trace in the next reference line generation.

In one possible embodiment, as shown in FIG. 5, step 106 comprises:

Step 1061, starting from the root node, according to the search direction and the preset search step, determine the next node from the first point cloud data.

After determining the search direction in the previous step, according to the preset search step, start the search for the next node from the root node P2, and the search direction is set to, assuming that the next node searched is P3, the next node searched satisfies the following formula:

（1）

Step 1062, determine whether the next node meets the target condition, the target condition is: the number of point clouds in the point cloud grid to which the node belongs is greater than or equal to the preset second quantity threshold; The second quantity threshold is greater than the first quantity threshold.

After searching for the next node from the first point cloud data according to the search direction and search step, in order to prevent the next node searched from being a noisy point cloud, the target condition can be used for verification.

Since most of the noise point clouds have been eliminated in the previous denoising processing steps, if there are a small number of noise point clouds in the first point cloud data, the amount of point cloud data near the location of these few noise point clouds will be very small. Therefore, the second quantity threshold can be set to determine whether the next node is a noisy point cloud according to the relationship between the number of point clouds in the point cloud grid to which the next node belongs and the second quantity threshold.

The second quantity threshold is greater than the first quantity threshold, for example, if the first quantity threshold is 1, the second quantity threshold can be set to 3.

In the step 104 denoising processing step, a smaller first quantity threshold is used to filter out the noise point cloud, and a larger second quantity threshold is not selected, in order to avoid the point cloud data on the lane line as noise point cloud rejection. In this way, collecting a small first quantity threshold may cause a small amount of noise point cloud to be mixed in the first point cloud data, so the second quantity threshold is used here to determine whether the next node is a noise point cloud. This step-by-step method of adding thresholds makes the scheme more feasible and applicable.

Step 1063, if the next node satisfies the target condition, the average of the coordinate values of each point cloud in the point cloud grid is determined, and the average value is used as the coordinate value of the new node.

If the next node, P3, meets the target condition, the point cloud grid to which the next node belongs is considered to be on the lane line.

If the next node is exactly at the edge of the lane line, it will lead to a large number of point clouds in the area that belongs to the lane line in this point cloud grid, and a small number of point clouds in the area that does not belong to the lane line, and the reference line is preferably located in the middle of the lane line. To avoid this deviation from the reference line, take the average of the coordinates of each point cloud in the point cloud grid as the coordinate value of the new node.

In one possible embodiment, after step 1062, further comprises:

Step 1064, if the next node does not meet the target conditions, the search step is increased, and the new next node is determined from the first point cloud data;

Step 1065, determine whether the new next node meets the target conditions;

Step 1066, if the target condition is not met, the search step is continued to increase, and the new next node is determined from the first point cloud data until the new next node meets the target condition, or the search reaches the preset stopping condition.

In steps 1064-1066, if the next node does not meet the target condition, the search step ψ is added to the original search direction, the next node P4 is recalculated, and whether the next node P4 meets the target condition.

If P4 meets the target condition, the average point cloud coordinates of the point cloud grid to which the next node P4 belongs is used as the coordinates of the new node. Thereafter, the acquisition timestamp of the new node is obtained, and in the case of the corresponding first trajectory data within the preset time range of the acquisition timestamp, based on the forward direction of the first trajectory data, the search direction of the next new node is determined, that is, the search direction is regenerated from step 105 for subsequent search steps.

If P4 does not meet the target condition, continue to increase the search step to search until the next node that meets the target condition.

If there are no point cloud meshes that meet the target criteria in the entire first point cloud data, the search is stopped. In this case, the first trajectory point T1 is updated, the trajectory point of the next moment corresponding to the first trajectory point (that is, the first rear point in the preceding scheme) is used as the central point, the pavement point cloud data within the preset radius of the first rear point is preloaded, and from step 104, the steps of denoising processing, determining the search direction and node search are re-executed.

In addition, during the search process, if a preset stop condition is encountered, the search is stopped.

In steps 1061-step 1066, in the above embodiment, starting from the root node, determine the target node according to the search direction and the preset search step, determine whether the target node meets the set target conditions, and finally determine the coordinate value of the new node, when the target node does not meet the set target conditions, the method of increasing the search step size is used to continue to determine whether there is a target node that meets the conditions, and if no target node that meets the conditions is found after increasing the search step twice, the trajectory point T1 is updated, The coordinate values of the new node are determined in the same way as described above, which improves the accuracy of the new node.

In one possible embodiment, the preset stopping conditions, comprising:

The new node has reached the intersection; The search direction of the new node is opposite to the direction of travel of the collection vehicle; The third distance between the new node and the nearest existing reference line start node is less than the third distance threshold; The existing reference line is the reference line of the same direction lane; The search step of the new node exceeds the preset step threshold.

In an embodiment of the present invention, if any of the above preset stopping conditions appear, the search for a new node is stopped.

Among them, if a new node has reached the intersection, indicating that the lane line has reached the endpoint, the search for the new node on the lane line can be stopped.

If the search direction of the new node is opposite to the driving direction of the collection vehicle, that is, the direction determined by the reference line selection node is opposite to the driving trajectory of the collection vehicle, it means that the search direction of the new node may be wrong and the search direction needs to be reset.

If the third distance between the new node and the start node of the existing reference line in the nearest codirectional lane is less than the third distance threshold, the reference line obtained by the current search can be merged into the existing reference line without the search for the new node. where the third distance threshold can be set to half the current lane width, for example, if the current lane width is 4 meters, the third distance threshold can be set to 2 meters.

If the distance between the new node and the starting node of the reference line of the codirectional lane is less than half of the width of the current lane, indicating that the new node belongs to the node on the co-directional lane, then the reference line obtained by the current search and the co-directional lane belong to the same reference line, and the reference line obtained by the current search can be merged into the reference line of the co-directional lane.

Specifically, the method of determining the reference line of the same direction lane is:

Update the new node to the root node, determine whether there is a drawn reference line near the root node, in the presence of a drawn reference line, set the initial node of the existing reference line to P', if there is an angle greater than or equal to 90° with , judge the existing reference line as the reference line of the opposite lane, and ignore the existing reference line. If the angle between and is less than 90°, the existing reference line is judged to be the reference line of the same direction lane.

If the angle with is less than 90°, the third distance between P2 and the nearest existing reference line start node can be obtained, the third distance threshold can be set to 2 meters, if the third distance is less than 2 meters, the last determined new node is set as the initial node of the existing reference line and the search for the new node is stopped.

In addition, if the search step of the new node exceeds the preset step threshold, the search step is too long and the error is too large, and the search needs to be stopped.

In the present embodiment, the step threshold may be set to 4 meters, in the process of searching for a new node, if the search step of the new node exceeds the preset step threshold, the node search is stopped.

In one possible embodiment, after step 108, further comprises:

Step 110, based on the position relationship between the lane reference line and each lane line, the preset lane width generates a plurality of lane lines of the road surface.

A lane guide is one of several lane lines, and each lane line is a parallel position relationship, and the lateral distance between other lane lines and them is determined according to the position of the lane guide. Then, by copying and panning the lane guides, additional lane lines can be generated.

FIG 6 is a schematic diagram of the structure of a lane reference line extraction device provided by an embodiment of the present invention. As shown in FIG. 6, the extraction device 200 of the lane reference line comprises:

Pavement point cloud data acquisition module 201, for obtaining pavement point cloud data, the pavement point cloud data is the point cloud data on the road surface collected by the collecting vehicle during driving, the pavement includes a plurality of lane lines;

The starting point root node acquisition module 202 is configured to obtain the starting point and the root node from the pavement point cloud data, the starting point and the root node are two points on the manually preconfirmed lane line; The direction from the starting point to the root node is consistent with the direction of travel of the collection vehicle; the manually preconfirmed lane line is one of the multiple lane lines;

the first trajectory point determination module 203, for obtaining the first trajectory data composed of the driving trajectory point of the acquisition vehicle, and determining the first trajectory point closest to the root node from the first trajectory data;

The denoising processing module 204 is configured to take the first trajectory point as the center point, preload the pavement point cloud data within the preset radius of the center point, and denoise the preloaded pavement point cloud data to obtain the first point cloud data;

The first search direction determination module 205 is configured to obtain the acquisition timestamp of the root node, and in the case of the corresponding first trajectory data within the preset time range of the acquisition timestamp, the search direction of the new node is determined based on the forward direction of the first trajectory point;

a new node search module 206, configured to start from the root node, according to the search direction and a preset search step, from the first point cloud data to search for the new node;

Update module 207 for updating the first trajectory point to the new node, starting with the new node, continuing to perform the denoising processing and search for the next new node operation;

Lane guide generation module 208, for extracting and generating lane reference lines based on the starting point, the root node and a plurality of new nodes searched.

Optionally, the first search direction determines module 205 specifically for:

Obtain two points adjacent to the first trajectory point from the first trajectory data to obtain the first pre-point and the first posterior point, the first pre-point is the trajectory point of the previous moment corresponding to the first trajectory point, and the first post-point is the trajectory point of the next moment corresponding to the first trajectory point;

obtain the angle between the first direction determined by the starting point and the root node, and the second direction determined by the first pre-point and the first post-point;

determine the first distance between the first trajectory point and the root node;

Based on the first distance, the angle determines the search direction of the new node.

Optionally, the first search direction determination module 205 is further used for:

If the first distance is greater than or equal to the preset first distance threshold, or the angle is greater than or equal to the preset angle threshold, the direction determined by the starting point and the root node is determined as the search direction of the new node; The first distance threshold is the distance traveled by the acquisition vehicle in P trajectory data acquisition cycles;

If the first distance is less than the preset first distance threshold, and the angle is less than the preset angle threshold, the search direction of the new node is determined based on the first trajectory point, the first front point, and the first back point.

Optionally, the first search direction determination module 205 is further used for:

determine the second distance between the first posterior point and the first trajectory point;

If the second distance is greater than or equal to the preset second distance threshold, the direction determined by the first pre-point and the first post-point is used as the search direction of the new node; The second distance threshold is the distance traveled by the acquisition vehicle in Q trajectory data acquisition cycles, Q>P;

If the second distance is less than the second distance threshold, the first trajectory point is updated to the first postpoint, and the direction determined by the pre and posterior points of the first postpoint is taken as the search direction of the new node.

Optionally, the lane reference line extraction device 200 further comprises:

The second search direction determination module is configured to use the direction determined by the starting point and the root node as the search direction of the new node in the absence of corresponding first trajectory data within the preset time range of the acquisition timestamp.

Optionally, the denoising processing module 204 is specifically configured for:

The preloaded pavement point cloud data is sorted by the point cloud reflection intensity;

The point cloud data sorted in the top N% and the point cloud data sorted in the last M% are filtered from the pavement point cloud data to obtain the primary processing point cloud data; The point cloud data sorted in the first N% is the point cloud data corresponding to the foreign object on the road surface, the point cloud data sorted in the last M% is the point cloud data corresponding to the ordinary road surface, and the initial processing point cloud data is the point cloud data corresponding to the lane line;

The spatial area corresponding to the initial processing point cloud data is divided according to the preset grid size to obtain a plurality of point cloud meshes;

According to the size relationship between the number of point clouds included in the point cloud grid and the preset first quantity threshold, each point cloud grid is initially binarized marked, and the plurality of point cloud grids are corroded and expanded according to the initial binarization marking results, and a plurality of point cloud meshes including target binarization marking results are obtained;

According to the target binarization labeling results, the target point cloud grid greater than or equal to the first quantity threshold is obtained, and the point cloud data in the target point cloud grid is determined to be the first point cloud data.

Optionally, the denoising processing module 204 is further used for:

The point cloud grid included in the point cloud grid where the number of point clouds is less than the first quantity threshold is marked as P, and the point cloud grid with the number of point clouds greater than or equal to the first quantity threshold is marked as Q to obtain the initial binary labeling result;

According to the initial binarization labeling results, the plurality of point cloud grids are subjected to multiple corrosion treatment and multiple expansion treatments, respectively, and a plurality of point cloud grids including the target binarization marking results are obtained;

The point cloud data in the point cloud grid with the target binarization mark result of Q is determined to be the first point cloud data.

Optionally, the new node search module 206 is specifically used for:

Starting from the root node, the next node is determined from the first point cloud data according to the search direction and the preset search step;

determine whether the next node satisfies the target condition, the target condition is: the number of point clouds in the point cloud grid to which the node belongs is greater than or equal to the preset second quantity threshold; the second quantity threshold is greater than the first quantity threshold;

If the next node satisfies the target condition, the average of the coordinate values of each point cloud in the point cloud grid is determined, and the average value is used as the coordinate value of the new node.

Optionally, the new node search module 206 is specifically used for:

If the next node does not meet the target conditions, the search step is increased to continue to determine the new next node from the first point cloud data;

determine whether the new next node satisfies the target condition;

If the target condition is not met, the search step is continued to increase, and the new next node is determined from the first point cloud data until the new next node meets the target condition, or the search reaches the preset stopping condition.

Optionally, the lane reference line extraction device 200 further comprises:

Lane line generation module for generating multiple lane lines of the road surface based on the position relationship between the reference line and each lane line, preset lane width.

Those skilled in the art can clearly understand that for the convenience and conciseness of the description, the specific working process of the extraction device of the lane reference line described above may refer to the corresponding process in the embodiment of the aforementioned method, which will not be repeated herein.

In yet another embodiment provided by the present invention, there is also an electronic device, the electronic device comprising a processor and a memory, the memory stored with at least one instruction or at least one program, at least one instruction or at least one program loaded by the processor and executed to achieve the lane reference line extraction method described in the embodiment of the present invention.

In yet another embodiment provided by the present invention, a computer-readable storage medium is also provided, the storage medium is stored with at least one instruction or at least one paragraph of program, the instruction or at least one paragraph of program is loaded by the processor and executed by the processor to achieve the lane reference line extraction method described in the embodiment of the present invention.

The above is mainly from the perspective of the device to provide an embodiment of the present invention provided by the scheme is introduced. It is understood that in order to achieve the above functions, the device comprises a hardware structure and/or software module corresponding to each function. Those skilled in the art should easily realize that in combination with the algorithmic steps of the examples described in the embodiments disclosed herein, the present invention may be implemented in hardware or a combination of hardware and computer software. Whether a function is performed in hardware or as computer software drives hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel may use different methods for each particular application to achieve the described function, but such implementation should not be considered beyond the scope of the present invention.

The above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited to this, and any change or replacement within the scope of the technology disclosed in the present invention should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.