Improved efficient bird’s eye view 3d detection boosts driving safety
Bi-Level Routing and Sparse Spatial Attention based Multi-View BEV 3D Object Detection for Autonomous Driving
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Detecting 3D objects around self-driving cars using overhead views is tough and slow because of complex calculations and combining information from many cameras. The authors created a new method called Sparse-BEVNet that uses special attention techniques to reduce the workload in processing images and combines features better without slowing down. Their approach also makes projecting camera views to a top-down view more efficient by skipping unnecessary calculations. This results in better accuracy detecting objects in driving scenes, as shown by tests on a standard dataset.
What this means in practice
- •For autonomous vehicle engineers: Implement a more efficient and accurate 3D object detection method using multi-camera views for safer self-driving systems.$Commercial implications: Enables improved perception software products for autonomous cars with higher accuracy and lower computation costs.
- •For robotics developers: Use the sparse attention techniques to enhance real-time 3D environment mapping from multiple cameras in mobile robots.
Authors
Jing Zhang, Jiaqi Liu, Zibo Wang
Abstract
Bird's Eye View (BEV)-based multi-view 3D object detection suffers from challenges of computational complexity, multi-scale feature extraction, and efficiency of dense 2D-to-BEV view transformation. To address these problems, this paper proposes an improved BEV 3D object detection algorithm Sparse-BEVNet. Firstly, a Bi-Level Routing Attention (BRA) mechanism is introduced into the image feature extraction network to reduce the computational burden of the backbone. Second, Cascaded Group Attention (CGA) is employed in the feature fusion module, which enhances deep interaction across features of different hierarchical levels without introducing additional computational overhead. Furthermore, a Sparse Spatial Cross-Attention mechanism is adopted to replace the conventional dense view projection pipeline. Experimental results on the public nuScenes dataset demonstrate that the proposed method achieves a mean Average Precision (mAP) of 45.2% and a nuScenes Detection Score (NDS) of 54.5%, corresponding to 3.6% and 2.8% improvements relative to the baseline model, respectively.