基于激光雷达与毫米波雷达融合的车辆目标检测算法

Vehicle target detection algorithm based on fusion of lidar and millimeter wave radar

针对仅基于单一传感器的目标检测算法存在检测精度不足及基于图像与激光雷达的多传感器融合算法检测速度较慢等问题,提出一种基于激光雷达与毫米波雷达融合的车辆目标检测算法,该算法充分利用激光雷达点云的深度信息和毫米波雷达输出确定目标的优势,采用量纲一化方法对点云做预处理并利用处理后的点云生成特征图,融合毫米波雷达数据生成感兴趣区域,设计了多任务分类回归网络实现车辆目标检测.在Nuscenes大型数据集上进行训练验证.结果表明:检测精度可达60.52%,每帧点云检测耗时为35 ms,本算法能满足智能驾驶车辆对车辆目标检测的准确性和实时性要求.

To solve the problems of insufficient detection accuracy of the target detection algorithm based on single sensor and slow detection speed of the multi-sensor fusion algorithm based on image and lidar,a vehicle target detection algorithm was proposed based on the fusion of lidar and millimeter wave radar.The depth information of lidar point cloud and the target information of millimeter wave radar were fully used,and the normalization method was adopted to preprocess the point cloud for generating feature maps.The millimeter wave radar data were fused to generate interested regions,and a multi-task classification regression network was designed to achieve vehicle target detection.The training verification was performed on Nuscenes large data set.The results show that the detection accuracy can reach 60.52%,and the point cloud detection takes 35 ms per frame.The proposed algorithm can meet the accuracy and real-time requirements of intelligent driving vehicles for vehicle target detection.