Multi-Scale Dilated Sparse Convolution for 3D Point Cloud Object Detection

With the rapid development of unmanned driving and intelligent transportation, 3D point cloud object detection methods have received widespread attention. Due to the disorder, sparsity, and unstructured characteristics of point clouds, building an effective point cloud object detection network and improving its accuracy become challenging. Therefore, multi-scale dilated sparse convolution(MSD) for 3D point cloud object detection is proposed, which utilizes multiple branches and convolutional kernels with different scales to capture feature information and improve object detection accuracy. The experiment on the KITTI dataset shows that this method further improves the accuracy of object detection, with the mAP (mean Average Precision) of 77.75%, demonstrating the superiority of this method.