Detection And Identification Of Road Traffic Signs Natural Environment

Traffic sign recognition is a key research direction of Intelligent Transportation System (ITS). It can be applied to unmanned vehicle and driver assistance system, and provides useful information for the automatic or semi-automatic driving. Through decades of research, the theoretical system and practice system in the field of traffic sign recognition gradually formed, and made a lot of breakthroughs.The research of our paper covers the detection, feature extraction and recognition of traffic sign.In the detection phase of traffic sign, according to the color and shape characteristics of traffic signs, this paper presented a method that based on color segmentation and local Hough transform. First, we did color segmentation to obtain a binary image that including traffic signs, and did a contour tracking on the binary image to extract candidate regions of traffic sign, then did a pre-classification of shape based on average RGB value of contour. And then used the Hough transform to do shape detection on candidate regions. Finally, located the traffic sign on the picture.During feature extraction and recognition, this paper studied the method based on the local KFDA. Those local methods that based on sub-pattern KFDA (Sp-KFDA) and module KFDA (MKFDA) were introduced in this paper. Then, according to the distribution of traffic signs’ information, we proposed a method that based on adaptively weighted module KFDA (Aw-MKFDA). Through those experiments on the classifier of K-nearest neighbors showed that the traffic sign recognition based on Aw-KFDA had a higher recognition rate than some other methods.Through a depth analysis of test samples that were identified wrong, we found that a higher degree of similarity of samples prone to misclassification. For solving this problem, we proposed a method that based on two phase traffic sign recognition of similar classes’ dividing. The method divides process of traffic sign into two phases. In the first phase, we used sparse representation to classify the test sample to the big class. In the second phase, we used sparse representation to classify the test sample to the specific class. In the process of using sparse representation for traffic sign recognition, we proposed a method that used local dictionary instead of global dictionary, to solve the problem of large dictionary that was caused by the large samples of traffic signs. Through some experiments indicated that our method could effectively improve the recognition of traffic sign.At last, we used the pictures that was from the real-time collection of unmanned platform to do experment, validated those methods of detection and recognition were proposed in our paper. Finally, we did a combination of previous methods and presented a recognition method based on sparse representation and local KFDA. Through some experiments indicated that our method got a very satisfactory result, and had a good stability.