Traffic Sign Intelligent Detection Method And Its Application In Road Property Management

Traffic management and traffic safety problems are more and more attentioned by people. In this background, the concept of intelligent transportation system arises at the historic moment. Intelligent detection of Traffic sign as a part of intelligent transportation system, plays an important role in intelligent management system for highway road information collection. Real traffic scene is complicated. Problems, such as conditions of light and weather, partial sheltered, obstruction from similar background or shadow, bring challenges to traffic sign detection and recognition. Therefore, researchers have been in pursuit of developing robust detection method of traffic sign. Aimed at the complicated situations, in this paper, our researches mainly focus on three aspects: haze remove of origin image, rapid location of traffic sign and semantic recognition of traffic sign.(1)This paper proposes a haze removal method based on normalized cut.Aimed at the haze origin image, this paper proposes a haze removal method based on normalized cut. On the base of the classic dark channel prior algorithm and normalized cut, this paper proposes a haze removal method. Results show that this method eliminates the halos generated by the classic dark channel prior algorithm, and improves the image quality. This algorithm, applied to road information collection system, can improve the precision of information collection for outdoor jobs in haze(2)This paper proposes a detection algorithm based on HOG feature of traffic sign with Fisher classifier.This paper uses fisher classifier trained by HOG features of traffic sins to achieved coarse detection of traffic signs. In this stage, algorithm does not require the precision, but as far as possible to reach 100% of recalls. This paper chooses GTSRB dataset as the training and testing sample set. The test results show that the algorithm has a good performance and a certainly robustness under most kinds of conditions.(3)This paper proposes a recognition algorithm based on HOG feature of traffic sign with SVM classifier.In view of the candidate ROI, this paper puts forward a fine-grained recognition algorithm combined with the SVM classifier. In fine-grained recognition stage, the classification process is divided into two levels of recognition. The first level can remove the ROIs which is not signs, so that it make sure the traffic sign recognition algorithm with high precision. The second level can output the semantics of the traffic sign. Experiments show that the recognition algorithm of twice classification has higher recall and precision, comparing with one-level classifier model.