Multi-camera Registration with Small FOV Based on Pose Graph Optimization

Multiple roadside cameras sense and localize autonomous vehicles for Automated Valet Parking. The accurate multi-camera registration into a shared coordinate system is of great importance. Owing to the sparse and distributed placement of multiple cameras, there is a very small field of view (FOV) between the cameras. The multi-camera registration is complex and dependent on human expertise. Therefore, a multi-camera registration method with small FOV based on pose graph optimization is proposed in this paper to automatically generate the extrinsic. The moving checkerboard in the public FOV is utilized as a reference for calibration. The optimization method constructs a pose graph to store camera and checkerboard pose and solves the pose graph by calculating checkerboard reprojection errors. To increase the overall flatness of the ground, ground plane assumptions can be utilized to further optimize the parameters. The experimental results show that the proposed method has greater calibration accuracy on the constructed multiple RGB-D camera system, as well as being simple and efficient in the real calibration procedure.