基于激光雷达的城轨交通障碍物检测算法研究

Obstacle detection algorithm of urban rail transit based on lidar

侵限障碍物对城市轨道交通有巨大的危害,现有基于通信的列控技术无法对其自动化防护。为避免障碍物侵入行车区间造成安全事故,利用激光雷达作为主要传感器,提出一种非接触式障碍物检测算法。借助激光雷达具有不受环境光照的影响,测距精度高,可实现远距离探测等优势,该算法能够实现对百米内障碍物的有效检出,同时具备高可靠性,不受隧道内光照条件差等恶劣条件的干扰。为克服隧道中坡度影响,结合点云校准算法将点云平面与地平面对齐。对隧道环境和站台环境建模分析,提出基于规则的轨道平面分割算法和基于区域增长的背景点云分割算法,有效实现地平面的分离以及背景点云的滤除。考虑到点云密度在不同距离分布不均,提出自适应欧式聚类障碍物检测算法。为验证整体算法的有效性,在宁波地铁5号线采集大量正线数据,进行障碍物注入仿真实验。实验结果表明:复杂运行场景下该障碍物检测算法在视距范围内低于70 m障碍物检出率可达85.89%,雷达超视距的情况下检出率有一定的衰减,低于120 m的障碍物检出率为63.08%。算法平均耗时为37.86 ms。

Obstacles intruding into the boundary pose great hazards to urban rail transit,and existing communication-based train control(CBTC)technologies are unable toprovide automated protection against them.To prevent safety accidents caused by the intrusion of obstacles into the driving area,a non-contact obstacle detection algorithm was proposed byusing lidar as the main sensor.With the advantages of high measurement accuracy,long-distance detection capability,and immunity to environmental light interference,the radar-based algorithm can effectively detect obstacles within 100 meters with high reliability,and is not affected by poor lighting conditions in tunnels or other adverse conditions.To overcome the influence of slope variation in the tunnel,a point cloud calibration algorithm was first used to align the data plane with the ground plane.Then,based on the modeling analysis of the tunnel environment and platform environment,a rule-based track plane segmentation algorithm and a region-growing background point cloud segmentation algorithm were proposed to effectively separate the ground plane and filter out the background point cloud.Finally,an adaptive Euclidean clustering obstacle detection algorithm was proposed to addressthe uneven distribution of point cloud density at different distances.To verify the effectiveness of the overall algorithm,a large amount ofmainline data was collected fromNingbo Metro Line 5 for obstacle injection simulation experiments.The experimental results show that under complex operating scenarios,the obstacle detection algorithm can achieve a detection rate of 85.89%for obstacles with a size smaller than 70 m within the visual range.In the case of exceeding the visual range,the detection rate decreases to 63.08%for obstacles smaller than 120 m.The average processing time of the algorithm is 37.86 ms.