Research on real-time 3D object detection and multi-task learning framework for autonomous driving scenarios

This paper proposes a multi-scale fusion attention-enhanced multi-task learning algorithm (MS-FA-MTL) framework for real-time 3D object detection in autonomous driving scenarios. The algorithm effectively integrates point cloud data features of different resolutions through a unique multi-scale feature fusion module, and uses the attention mechanism to enhance the focus on key target areas. The results show that compared with the traditional method, the MS-FA-MTL algorithm has improved the detection accuracy by 18% to 93.5%, shortened the average detection time by 30ms, and achieved a real-time detection speed of 45 frames per second. In terms of multitask performance, the target classification accuracy reached 91.2%, and the average error of distance estimation was reduced by 1.2 meters, which is significantly better than the comparison algorithm, providing strong support for the safe and reliable operation of the autonomous driving system.