A Two-Stage Pillar Feature-Encoding Network for Pillar-Based 3D Object Detection

Three-dimensional object detection plays a vital role in the field of environment perception autonomous driving, and its results are crucial for subsequent processes. Pillar-based 3D is method to detect objects by dividing point cloud data into pillars extracting features from each pillar. However, current pillar-based object-detection methods suffer problems such as “under-segmentation” false detections overlapping occluded scenes. To address these challenges, we propose an improved network with two-stage pillar feature-encoding (Ts-PFE) module that considers both inter- intra-relational among pillars. This novel approach enhances model’s ability identify local structure global distribution data, which improves distinction between scenes ultimately reduces under-segmentation problems. Furthermore, use attention mechanism improve backbone make it focus on important features. The proposed evaluated KITTI dataset. experimental show accuracy significantly benchmarks BEV 3D. improvement AP car, pedestrian, cyclist 1.1%, 3.78%, 2.23% over PointPillars.