Multi-view Fusion Based Object Detection Approach Using Dempster-Shafer Evidence Theory

LiDAR, as a key sensor in the field of 3D object detection, requires substantial computational resources for object detection based on point cloud data. However, reducing data dimensionality through a single projection view may result in information loss. To address the issues of false detections and missed detections in single-view object detection, a multi-view fusion based object detection approach using the DempsterShafer (D-S) evidence theory is proposed. The point cloud is projected and encoded into multi-channel $2 D$ range views and bird’s-eye views. Subsequently, the encoded bird’s-eye view and range view are respectively fed into two different networks to perform object detection. Finally, the D-S evidence theory is utilized to fuse object detection results from these two views, yielding the final object detection outcome.