Polyline-driven stop sign detection
Qi Li, Yongyi Gong · 2016
Exhaustive scanning is a popular scheme for the detection of visual objects in images. In this paper, we propose a polyline-driven detection scheme with an application to stop sign detection. Given an input image, we first extract basic polylines, including line segments and 2-piece polylines, from its edge image. Line segments are then used to generate a set of hypothesis boxes, i.e., a space of candidates, according to the orientations of line segments and the geometry of a regular octagon. To compute the confidence scores of candidates, we propose a design of eight voting districts of a hypothesis box and a score assignment scheme that integrates four basic features of a 2-piece polyline. We test the proposed method on a dataset that contains 500 stop sign images, and obtain a result of 97% detection rate.