Real-time Target Recognition for Urban Autonomous Vehicles Based on Information Fusion

Abstract

Aiming at the problem that the single sensor has insufficient sensing dimensions and poor real-time performance, a real-time target recognition method for urban autonomous vehicles based on the fusion of lidar and camera is proposed. To achieve pixel-level matching of the two sensors, a coordinate transformation model between the two sensors is established; the yolov3-tiny algorithm is improved to increase the accuracy of target detection. Voxel grid filtering was performed on the lidar points, the ground is filtered according to the lidar point slope; the model of clustering radius and distance is established, and non-ground point clouds are clustered; the idea of envelope in images is introduced to obtain the 3D bounding box and pose information of the target; the visual target features are fused with the lidar target features. The experimental results show that the improved yolov3-tiny algorithm has a higher recognition rate for dense urban targets. The lidar algorithm can complete three-dimensional target detection and pose estimation. The fusion recognition system meets the actual driving requirements in terms of accuracy and real-time performance.