Human Joints Auto-Calibration Method Based on SuperGlue

Camera calibration is a fundamental and crucial issue in computer vision and image processing. It entails obtaining the camera's internal and external parameters. Nevertheless, the current widely used calibration methods rely on the usage of calibrators, which is time-consuming and inconvenient. A high-precision multi-camera relative posture estimation approach based on human joints is suggested in this paper. Without using specialized calibration tools like checkerboards, only one person needs to enter the calibration scene to calibrate the camera. Firstly, the traditional feature matching approach is replaced by a novel neural network called SuperGlue, which is based on a graph convolutional neural network and an attention-based flexible contextual aggregation mechanism. Secondly, the human joints detection algorithm has been upgraded compared to our earlier work. Finally, the internal and external parameters of the camera are iteratively optimized to obtain the optimal camera parameters. According to the experiment results, the calibration efficiency of the multi -camera system is improved and can be applied in many computer vision tasks. This approach greatly shortens the calibration procedure's time compared to Zhang's calibration method.