Abstract
Current 3D object detection technology for autonomous driving usually cannot efficiently utilize local sensitive points. Meanwhile, contextual feature extracted from a object is not sufficient, which easily leads to deteriorated detection accuracy of the final object estimation. For the problems, a point-voxel-based 3D dynamic object detection algorithm is proposed. Firstly, local points are grouped with a camera frustum. Then the global feature extracted by the submanifold 3D voxel CNNs is aggregated into frustum key points. Secondly, a module of vector pool with feature aggregation is used to aggregate multi-scale features of the point cloud. Moreover, the frustum raw feature and BEV feature are used for feature extension. Subsequently, the fine multi-scale feature extracted from the point cloud is used as input to a subsequent fully convolutional network for final classification and continuous estimation of oriented 3D boxes. The proposed method was compared with other state-of-the-art algorithms on the KITTI, Waymo, and nuScenes datasets. Experimental results showed that the proposed algorithm was better in accuracy, robustness, and generalization capabilities in 3D dynamic object detection. Experiments on a real scenario and extensive ablation studies also demonstrated that the proposed algorithm not only effectively controls computational cost, but also achieved more efficient results in 3D object detection.
Funding Information
  • National Natural Science Foundation of China (61801323,61972454,62201375)
  • China Postdoctoral Science Foundation (2021M691848)
  • Natural Science Foundation of Jiangsu Province (BK20201405,BK20220635)
  • Science and Technology Projects Fund of Suzhou (SS2019029, SYG202142)