应用于相机标定的亚像素棋盘角点检测OA

In the camera calibration process,the accuracy of chessboard corner detection is crucial for ensuring the precision of calibration results.Addressing the insufficient accuracy of current chessboard corner detection methods,this study proposes a no-vel sub-pixel level chessboard corner detection technique.Firstly,the U-Net convolutional neural network is employed as the backbone network to construct corner heatmaps based on the captured chessboard images.In addition,in order to reduce the se-mantic gap between encoder and decoder features,this research innovatively introduces a channel and spatial dual-cross attention module.Subsequently,precise sub-pixel chessboard corner coordinates are calculated using a Gaussian surface fitting method.Experimental results demonstrate that this method effectively improves the accuracy of corner detection and achieves lower repro-jection error in camera calibration tasks.