A new straight edge detection algorithm using direction-controlled edge tracking and random hitting

This paper presents a new efficient straight edge detection algorithm for autonomous robot navigation and other visual tasks. First, we propose a new edge tracking algorithm that gives the endpoints of the detected edge. To obtain robust edge directions we use a least-squares linear fitting method to control the tracking process. Based on this “direction-controlled” edge tracking method, we construct an algorithm to find all major edges in an image. To minimize the exhaustive pixel processing, we adopt a random hitting method using a pseudo random number generator. Only if an initial pixel generated is significant, we track the edge that the pixel belongs to. To avoid finding the same segment twice, we examine whether a newly generated pixel is close to any of the already detected segments. Through our experiments we confirmed that most of major edges are detected by processing of an extremely small fraction of the total pixels. Only by processing of 4.36% of the total pixels we were able to detect 20 major line segments in a test image. Therefore, implementing this algorithm will be extremely effective in image understanding and visual applications including the model-based real-time robot navigation task