考虑路侧激光雷达摆动的背景建模及动态更新方法

Background Modeling and Dynamic Updating Method Considering Roadside LiDAR Swing

针对路侧激光雷达摆动和背景动态变化导致背景提取困难的问题,提出了一种考虑路侧激光雷达摆动的背景建模及动态更新方法。根据点云纵坐标构建点云高度矩阵并进行时域中值滤波构建背景模型,计算目标点云与背景模型的高度差,提出一种动态阈值选取策略实现对背景点云的滤除。此外,基于点云高度矩阵帧差实现背景模型的更新。对比试验结果表明,所提出的方法背景点滤除率保持在98%以上,目标点提取率保持在93%以上,所提出的方法相比于现有的方法具有更高的准确率和稳定性,单帧点云的处理时间约为32 ms,可满足智能交通系统的实时性需求。

For the difficulty of background extraction caused by roadside LiDAR swing and background dynamic change,this paper proposed a background modeling and dynamic update method considering roadside LiDAR swing.The point cloud height matrix was constructed by using point cloud ordinates,and the background model was constructed by median filtering in time domain.The height difference between the target point cloud and the background model was calculated,and a dynamic threshold selection strategy was proposed to filter the background point cloud.In addition,the background model was updated based on the frame difference of point cloud height matrix.The comparison test results indicate that,the rate of background points filtering of the proposed method is above 98%,and the rate of target point extraction is above 93%.Compared with the existing background filtering methods,the proposed method has higher accuracy and stability.The processing time of single frame point cloud is about 32 ms,which can meet the real-time requirement of intelligent transportation system.