An intra-image tracking algorithm for traffic sign recognition

Segmentation is a crucial task in an automatic traffic sign recognition system. In order to increase the success ratio detection, we propose two alternatives. Firstly, most traffic signs are designed in such a way that is easy to implement a double detection for each segmentation method. For instance, many european triangular and circular signs can be detected twice because of the red outer border and the white inner area. On the other hand, we propose a detection subsystem which includes several segmentations that can work in parallel for each image. However, even though a sign can be detected several times, the final result at the output system must give a single stroke for each sign. For this purpose, after the recognition process, the intra-image tracking unifies detections and establishes correspondences according to criteria based on color, geometric shape and position. For each incoming image a new track is initiated an all identified signs are compared, existing a structure that contains the tracked objects. If the recognition results are not the same for the different detections of a sign, the tracking algorithm decision is given by majoritary vote.