Point Cloud Object Detection Algorithm Based on Attention Gated Graph Convolution

Abstract

By referring to the behavior characteristics of driver attention allocation mechanism, it is expected to effectively improve the detection performance of laser point cloud perception algorithm for small size targets and occluded targets in complex scenes. In this paper, the attention-gated 3D object detection method of image accumulation point cloud is studied to achieve accurate perception of point cloud objects in complex scenes. Firstly, the attention gating mechanism is introduced to construct the point cloud map structure, strengthen the interaction and update between point cloud semantics and feature extraction network, and improve the feature extraction ability of the network for point cloud targets. Then, the most discriminative key nodes of the point cloud map are obtained by selecting the best pooling ratio coefficient to improve the classification performance of the algorithm. Finally, a variety of complex scenarios under the real vehicle test. The experimental results show that the proposed point cloud target detection algorithm based on attention gated graph convolution can effectively improve the operation efficiency and accuracy of the algorithm, and improve the intelligent level of the environmental perception system.