Feature detection and matching for traffic sign images

It is important to detect and recognize the traffic sign for mobile robot localization and navigation. In this paper, an algorithm frame of feature detection and matching has been developed which includes shape detection, Harris corner detection, SIFT feature matching and robust estimation method. Firstly, the color threshold segmentation algorithm in RGB color space is adopted to get the candidate region of traffic signs and the region growing method is applied to remove the noise in this image. Secondly, the shape features on the edge image are detected using template matching. Thirdly, Harris corner features are calculated and sorted, then the SIFT feature descriptors are computed on the extraction corner points. Finally, according to the minimum Euclidean distance the matching characteristic vectors are obtained between two images, then random sampling algorithm with robust estimation is used to reduce mismatch. Experiment result shows that this algorithm is efficient.