Automatic Traffic Sign Detection And Recognition

Advanced driver assistance systems(ADAS) refers to a variety of high-tech transportation system, the purpose of these systems is designed to help drivers improve aware of the potential threat to enhance road safety. As one of the important subsystems of ADAS, implementation of traffic sign recognition has become a challenging issue, and has become an important hot research topic in the field of intelligent transportation. Real-time traffic sign recognition systems can generally be divided into three phases: detection, tracking, and classification.Recently, studies have shown that methods using feature extraction can get better performance. In this paper, we take traffic signs in natural scenes as the research object, and mainly investigate image preprocess, traffic sign detection and sign recognition. The details of our work are as follows:1) In natural scenes, the existence of noise usually effect traffic sign detection. In terms of this problem, considering spatial correlation between pixels in local area, this paper exploits image smoothing preprocessing techniques to remove noises for sequent robust detection. Besides, this paper mainly investigates three preprocessing techniques: mean filter, Gaussian and median filter. Experiment shows the Gaussian filter is able to get better detection performance.2) Conventional segment algorithm based on mean shift can not get fine traffic sign candidate areas. In terms of this problem, based on traffic segment algorithm, this paper proposes an improved segment algorithm based mean shift, and finally the proposed method obtains robust detection performance. Experiment shows the improved mean shift algorithm is superior to traditional mean shift for detection performance.3) Traditional recognition methods based on low-level visual features can not accurately identify traffic signs. Firstly, we extract HOG features from traffic signcandidate areas, and then exploit SVM classifier to train these extracted HOG features for recognition, finally build a HOG-SVM traffic sign recognition model. Experiment on test datasets shows the proposed method can effectively obtain robust recognition performance.