Abstract: Intersection is a complex traffic scene in the development of autonomous driving. The cost of using high-definition map scheme is high, and it is difficult to effectively identify intersection shape only by vehicle sensors. Therefore, an intersection local path planning algorithm based on open source topology map and vision is proposed. First, uses the open source topology map to plan the global navigation path which was used as a guide line. Then, semantic segmentation technology is used to identify the freespace in the visual image. At the same time, the local path’s starting point, end point and control points are determined by combining the real-time positioning information of the vehicle at the intersection. Finally, bazel curve interpolation method is used to obtain the curve cluster of alternative paths, and the optimal local path is selected according to the result of the multi-dimensional weighted cost function. Then realizes the autonomous driving of the vehicle in the process of turning at the intersection. Experimental results show that the proposed strategy can effectively plan local paths at intersections without relying on high precision maps and improve the ability of autonomous vehicle to pass at intersections. The passing rate at intersection can reach 99%.

Key words: Autonomous vehicles; Local path planning; Intersection; Topology map; Computer vision

摘要： 交叉路口是自动驾驶开发过程中面临的复杂交通场景，采用高精度地图方案成本高昂， 而仅通过车载传感器难以有效识别路口形状，因此，提出了一种基于开源拓扑地图与视觉可 行驶区域检测技术的路口局部路径规划算法。首先，基于开源拓扑地图采用 A*算法规划全局 导航路径作为引导线，然后通过语义分割技术识别当前可行驶区域，并结合车辆实时定位信 息，在路口确定局部路径的起点、终点与一组备选控制点，最后采用贝塞尔曲线插值方法， 得到备选路径的曲线簇，根据多维度加权代价函数结果选取最优局部路径，进而实现车辆在 路口转弯过程的自动驾驶。实验结果表明，该策略能够在不依赖高精地图的情况下，在路口 处有效规划出局部路径，提高自动驾驶车辆在路口处的通过能力，路口通过率可达 99%。该 策略不依赖高精地图和激光雷达，对于自动驾驶量产降本具有重大意义。

关键词: 自动驾驶；局部路径规划；交叉路口；拓扑地图；计算机视觉