Multi-sensor 3D Target Detection Method with Fusion of Point Cloud Enhancement
Three dimensional object detection is an indispensable function module in automatic driving environment perception.The rich image texture information of camera sensor can make up for the sparse problem of laser radar point cloud.A plug and play RI Fusion module is proposed to achieve the effective fusion of lidar and camera.The point cloud is converted into a compact range view representation through spherical coordinate transformation,the distance images are integrated with the corresponding camera images based on the attention mechanism,the original distance images are connected with the fusion features to retain the information of the point cloud,and the fusion results are projected into the spatial point cloud.The point cloud after feature enhancement can be input into the 3D target detector based on lidar.The experiments are carried out on KITTI 3D target detection benchmark,the results show that the proposed fusion method can significantly enhance multiple 3D target detectors based on lidar,and can achieve higher detection accuracy for small targets such as pedestrians and cyclists,etc.
国家科技期刊平台
NSTL
万方数据
