摘要
为了保证智轨电车的运行安全,提高结构化道路环境下多种路况障碍物检测与跟踪的高效性和鲁棒性,文章提出一种基于激光雷达与毫米波雷达融合的多目标检测与跟踪方法。该方法首先基于毫米波雷达与激光雷达进行单传感器感知,然后通过多源融合算法融合单传感器感知结果,以获取更精确的感知结果。基于虚警滤除算法解决了毫米波雷达虚警引起误报问题,基于点云深度图和梯度与距离特征解决了多变场景下的障碍物聚类问题,基于分区点云特点设计代价函数提高了形状估计算法的鲁棒性,基于航迹信息进行多目标跟踪提高了目标跟踪的稳定性。为了克服异构传感器融合难题,设计了多源传感器异步融合策略,提升了结构化道路环境下目标检测与跟踪能力。最后,基于ROS框架在智轨电车上进行实车测试,试验结果表明,该方法能够稳定可靠地检测目标并实现跟踪。
To ensure the operational safety of the autonomous-rail rapid transit(ART) trams on the structured roads, it is necessary to improve the efficiency and robustness of obstacle detection and tracking under different road conditions. This paper proposed a multi-target detection and tracking approach based on the fusion of LiDAR and millimeter-wave radar. Firstly, a millimeter-wave radar and LiDAR are used to perceive by single sensor, and the sensing results were merged by using a multi-source fusion algorithm, to generate more accurate detection results. Based on the false alarm filtering algorithm, the problem of false alarm probably caused by the millimeter wave radar was solved. The approach was based on the point cloud range image, and the gradient and distance characteristics to solve the problem of clustering obstacles in changing scenes. The cost function was derived from the characteristics of the segmented point clouds, which improves the robustness of the shape estimation algorithm. The multi-target tracking based on the track information raised the stability of target tracking. To overcome the challenge of heterogeneous sensor fusion, a multi-source sensor asynchronous fusion strategy was applied,which improved the target detection and tracking capabilities on the structured roads. The proposed approach was verified on the real ART trams based on the ROS framework. The test results demonstrate the stability, accuracy and reliability of the proposed approach.
作者
潘文波
袁希文
林军
谢国涛
龙腾
黄文宇
PAN Wenbo;YUAN Xiwen;LIN Jun;XIE Guotao;LONG Teng;HUANG Wenyu(CRRC Zhuzhou Institute Co.,Ltd.,Zhuzhou,Hunan 412001,China;State Key Laboratory of Advanced Design and Manufacturing for Vehicle Body,Hunan University,Changsha,Hunan 410082,China)
出处
《机车电传动》
北大核心
2022年第4期157-165,共9页
Electric Drive for Locomotives
基金
国家自然科学青年基金项目(5210120159)
湖南省自然科学青年基金项目(2020JJ5091)。
关键词
智轨电车
分割聚类
形状估计
交互式多模型
多源信息融合
autonomous-rail rapid transit(ART)tram
segmentation clustering
shape estimation
interactive multi-model
multi-source information fusion
作者简介
通信作者:潘文波(1986-),男,博士,高级工程师,主要研究方向为智能驾驶环境感知与建图定位,E-mail:panwb1@csrzic.com。
