3D Object Detection for Multi-sensor Fusion Based on BEV Perspective
3D object detection is an important part in the road environment perception of autonomous driving.Existing mainstream framework is to obtain multi-modal data by using multiple sensing devices,which achieves multi-sensor fusion.There are the shorta-ges of geometric distortions and unequal information priorities in the fusion process of traditional cameras and LiDARs,resulting in in-sufficient 3D object detection performance of sensor fusion.To address this issue,a multi-sensor fusion 3D object detection algorithm based on bird's-eye view(BEV)is proposed.The lift-splat-shot(LSS)method is used to obtain the potential depth distribution of the image,and establish the feature map of the image in the BEV space.The set abstraction method of point-voxel region convolutional neural networks(PV-RCNN)is used to establish the feature map of the point cloud in the BEV space.A low-complexity feature en-coding network is designed for fusing multi-modal features in a unified BEV space in the proposed method to achieve 3D object detec-tion.Experimental results show that the proposed method improves the accuracy by 4.8％compared to the LiDAR method,reduces the parameters by 47％compared to the traditional fusion methods,and maintains similar accuracy.The proposed method meets the detection requirements of the road environment perception of autonomous driving system.
3D object detectionBEV viewmulti-sensor fusionautonomous drivingroad environment perception
万方数据
维普
