Three-dimensional reconstruction system aided by image calibration using color ICP based on graph optimization

The 3D reconstruction with multiple RGBD sensors is favored for its excellent real-time performance and high accuracy, where the performance largely depends on the precision of calibration among RGBD sensors. This paper proposes a graph-optimization-based calibration method for multiple RGBD sensors. Initially, a coarse calibration is assisted by the Zhang's calibration method; based on this, a fine calibration is further conducted through ColorICP to correct the extrinsic parameter estimation between each pair of sensors. Concurrently, this method computes the information matrix of the extrinsic parameter estimation. Utilizing both the extrinsic parameter estimation and the information matrix, a pose graph is constructed for global optimization estimation. The efficacy of this scheme in enhancing the accuracy and robustness of multi-RGBD calibration is demonstrated through applications in a gazebo simulation environment and a physical 3D reconstruction system built with four Intel Realsense L515 cameras, showcasing the potential applications of multi-RGBD sensor 3D reconstruction technology.