Traffic sign recognition and tracking for a vision-based autonomous vehicle using optimally selected features

Recently, developing a vision-based autonomous navigating vehicle system achieves more attention from many researchers. It is because the vision sensor provides a lot of information and is low-cost device rather than other sensors. Traffic signs, as one of the most important visual information, carry a lot of useful information required for navigation. Therefore, this paper addresses a framework for traffic sign recognition and tracking. First, the candidate region of traffic sign is localized using maximally extremal stable region on normalized blue and red color image. Second, scalable histogram of oriented gradient features are extracted from candidate region. Third, in order to obtain an efficient computation in the recognition stage, a feature selection method based on genetic algorithm is performed. Next, k-nearest cluster neighbor classifier is then processed to recognize region into a certain traffic sign class. Lastly, detection-based tracking is performed on consecutive frames for maintaining stable information of traffic sign. This information is then used by vehicle for navigation purpose. The extensive experiment was carried out over German traffic sign recognition database and video. The experimental results demonstrate the effectiveness of our systems. It is shown that the GA based approach enables to reduce feature size to 90% while maintaining the recognition performance. Also the tracking mechanism achieve reduction rate up to 50% without losing the true rate.