Research on fusion-based deep convolutional networks in 3D object detection
Yong Zhang, Chouchou Yue · International Conference on Electronic Information Technology (EIT 2022) · 2022
Aiming at the problem that the uniformly sampled point cloud is used for training in the three-dimensional object detection of RGB-D data, the accuracy of the point cloud in the actual scene is reduced, and the PointConv network is proposed to detect the loss of some key point cloud features. Firstly, by generating a 2D area proposal from the image and positioning it in the 3D point cloud data. Secondly, the 2D-driven 3D target detection is realized by the segmentation network based on PointConv and the 3D bounding box evaluation network. At last, evaluated on the KITTI and SUN RGB-D datasets, this method has real-time capabilities and surpasses the current detection levels of DoBEM, MV3D, Mono3D, Frustum PointNets ,and other methods. Experimental results show that compared with the previous state-of-the- art technology, our method improves mAP by 0.5% to 12.4%, and is 1-3 orders of magnitude faster.