A Corner Detection Method for Noisy Checkerboard Images.

This article proposes a novel approach for corner detection in noisy checkerboard images, comprising several methodical steps: (1) an initial extraction of corners utilizing the cross features present in the edge image of the checkerboard; (2) the elimination of erroneous corners through an analysis of the periodic consistency among the detected corners; (3) the identification of the outermost corners and the subsequent generation of a rectangular bounding box based on the total number of input checkerboard corners; (4) the reconstruction of missing corners, which may have been obscured by noise, by leveraging the periodic characteristics of the corners. Experimental findings indicate that this methodology is capable of effectively detecting all corners of the checkerboard across varying levels of noise, thereby significantly enhancing the success rate of corner detection in noisy images. This makes the proposed method particularly advantageous for camera calibration in special scenarios where noise or contamination in checkerboard images is unavoidable.