Image corner detection of CPDA-based on edge contour

Abstract

Coupled with slow detection speed, the existing corner detection methods based on chord-to-point distance accumulation (CPDA) can not accurately detect adjacent corners, a fast and accurate CPDA based corner detection algorithm was proposed. Firstly, comparative experiments were carried out to investigate the curvature threshold, angle threshold and chord length, and a chord was selected to accurately detect the chord length of adjacent corners. Secondly, the edge was detected by Canny edge detector, and then the chord length was used for curvature estimation to get the curvature of each pixel on the curve, and the curvature threshold was used to determine whether it is a candidate corner. Finally, the error detection point was removed by the angle threshold, and the final corner point was obtained. The experimental results show that the proposed detector has good corner location performance compared with CPDA and other three classical detectors. For the Checkerboard diagram, the number of correct corners detected by the detector of this algorithm is 49, which is similar to that of the standard detection. For the Lab diagram, the proposed detector has the best corner location performance, and its location error is only 1.477. Compared with FAST, GCM and CPDA detectors, this detector also achieves the best performance in terms of missed detection points, error detection points and detection speed.

Keywords

• corner detection
• chord-to-point distance accumulation
• edge contour
• curvature
• offset