A New Method for Polygon Detection Based on Hough Parameter Space and USAN Region
Shupei Li, Hui Zhang, Zhang Zhisheng, Xia Zhijie · 2020
This paper propose a new approach that combine Hough Transform (HT) and corner detection to detect polygons, which consider integrated characteristics not the individual characteristics. We establish a Polygon Parameter Space (PPS) to fit and characterize polygons, which consist of angles, coordinates, USAN values and every two lines of intersections. Firstly, canny operator is used to extract edges map, applied HT to detect line along edges of polygon shape and compute PPS. Secondly, corner detection among intersections is realized by comparing USAN value with angle of intersections, an adaptive threshold and adjusted brightness of nucleus of USAN is introduced to obtain accurate vertices from corners. Finally, we propose an algorithm based on Deep First Search (DFS) to fit the set of vertices regardless convex polygons (CVPs) or concave polygons (CCPs) according to parameters in PPS. The experimental results show that the proposed approach can effectively detect polygons with a less running time and higher accuracy, and shows the advantage of detecting the CVP and CCP shapes of broken vertices.