Object Detection from Point Clouds for Driverless Vehicles

Abstract Point clouds offer the advantage of providing direct access to 3D data. So, utilizing solely point clouds, we present an object detection technique for driverless vehicles in this study. In method, a graph is first constructed using k-nearest neighbour (KNN) and then neural network module proposed local information extraction. Then, depending on coordinates points, utilize pillar-based projection approach project locally informative feature into bird’s-eye-view (BEV). After that, residual-based networks with attention mechanisms are used BEV features processing. The system capable assigning appropriate weights several nearby hence improving performance. work achieves 58.49 percent 67.90 mAP (mean Average Precision) Object detection, respectively, KITTI benchmark. We also tested method our vehicle campus, compared it prior utilized. experimental results reveal that outperforms existing methods (about 2% higher than PointPillars detection) gives more accurate information.