Multi-Scale Feature Fusion 3d Vehicle Detection Based on Dual-Branch Convolution

Point cloud data contains rich 3D spatial information and is crucial for the perception systems of autonomous vehicles. Vehicle detection based on point cloud data is one of the important tasks of autonomous driving. In large-scale autonomous driving application scenarios, vehicle detection methods based on point cloud data still face the challenges due to the weak discriminability of object features. In this work we propose a multi-scale feature fusion 3D vehicle detection method based on dual-branch convolution. Specifically, a multi-scale feature fusion network based on dual-branch convolution is constructed to capture object details and contextual information more efficiently and suppress background noise interference. Additionally, we adopt the coordinate attention module to focus on key information of the fused feature and enhance the object discriminability. Experiments are conducted on the DAIR-V2X-I dataset and the KITTI dataset to evaluate the proposed method. The experimental results demonstrate that our approach significantly improves vehicle detection accuracy and achieve better performance than other state-of-the-art methods.