A lane line tracking method based on graph optimization comprises the following steps that S1, a positioning module provides vehicle positioning information, lane line points detected on an image are projected to the BEV of a vehicle body coordinate system, correlation matching is conducted on current detectlanes and historylanes through an SORT tracking algorithm, tracklanes are generated, and the tracklanes comprise current observation points and historical observation points; and creating nodes according to the observed states or variables, creating edges according to the constraint relationship among the observed states or variables, and generating a graph through the created nodes and edges. And S2, optimizing the graph by adopting a gradient descent method to minimize the target function of each edge in the graph, and further obtaining a final lane line parameter. According to the invention, various variables such as image domain information, information under a vehicle body coordinate system, vehicle attitude and the like are integrated, and the lane line tracking method is systematically designed.
一种基于图优化的车道线跟踪方法,包括如下步骤,步骤S1.定位模块提供车辆定位信息,将图像上检测到的车道线点投影到车体坐标系的BEV下,通过SORT跟踪算法,对当前detect_lanes和history_lanes做关联匹配,生成track_lanes,track_lanes包含当前观测点和历史观测点;根据观测的状态或变量创建节点,根据观测到的各个状态或者变量之间的约束关系创建边,通过创建的节点和边生成图。步骤S2.采用梯度下降法优化图,使图中每条边的目标函数最小化,进而得到最终的车道线参数。本发明把图像域信息、车体坐标系下的信息、车子姿态等各种变量综合起来,系统性地设计车道线跟踪方法。
Lane line tracking method based on graph optimization
一种基于图优化的车道线跟踪方法
2024-06-25
Patent
Electronic Resource
Chinese