Multi-modal Feature Fusion 3D Object Detection

Abstract

For the existing 3D small object detection is prone to false detection and missed detection and other deficiencies. A 3D object detection method based on multi-modal feature fusion is proposed. Firstly, a feature extraction module is designed. The input image data is down-sampled through the image feature extraction network, and the input point cloud data is sampled and grouped through the point cloud feature extraction network to obtain the feature information at different scales. Secondly, a multi-modal feature fusion module is constructed to realize the point correspondence between point cloud features and image features by projection operation, and then the image features and point cloud features are splicing and fused to generate the final point cloud features to compensate the deficiency of single modal feature information. The experimental results show that compared with the existing algorithms, the algorithm in this paper improves the average detection accuracy of small object by 2.03%.