【摘要】 为了保证智轨电车的运行安全,提高道路入侵物识别的高效性和鲁棒性,文章提出一种基于多激光雷达的道路入侵物检测技术。其利用ROS框架对多个雷达数据进行采集、融合,用于获得无重叠区域的低复杂度点云数据;采用Ground Plane Fitting算法,通过多关键点迭代多次生成模型来对点云数据进行去噪处理,返回低噪、易分割的点云数据;采用Scan Line Run算法,结合点云数据3D结构特点,从上而下进行无重复性的标签化处理,返回道路入侵物的位置、大小、点数等信息;将几种常见的道路入侵物进行特征提取并训练,基于之前标签化的入侵物进行分类处理。实验结果表明,该方法在实车运行中识别、分类均具有良好的鲁棒性和准确性。更多还原

【Abstract】 In order to ensure the operation safety of autonomous-rail rapid tram and improve the efficiency and robustness of road intrusion detection, a road intrusion detection technology based on multi-LiDAR was proposed in this paper. ROS framework is adopted to collect and fuse multiple radar data in order to obtain low complexity point cloud data without overlapping region. Point cloud data is de-noised by multi-key point iteration and multi-generation model using Ground Plane Fitting algorithm, and point cloud data with low noise and easy segmentation is returned. By using Scan Line Run algorithm and 3 D structure of point cloud data, non-repetitive tagging is performed from top to bottom to return the location, size and points of road intrusions. Several common road intrusions are extracted and trained, and classified based on the previously tagged intrusions. Experimental results show that the proposed method has good robustness and accuracy in recognition and classification.更多还原