Research On Methods Of Traffic Signs’ Detection And Recognition

Traffic signs’detection and recognition is an important part of the intelligent transportation system. It also has been a increasingly concerned research topic. This paper discusses the detection and recognition of road traffic signs in the real scene. Then summarize and present some methods of detection and identification. The detection and recognition of traffic sign generally includes steps of image preprocessing, image segmentation, location and classification.Because of the complex and varied condition, images will be better segmented after preprocessed. Here the paper applies Gamma correction on R, G, B channels of images directly. Using a look-up table to improve efficiency of preprocessing. Adding the white threshold and black threshold into Gamma Correction makes the processed image not only the brightness but also the saturation and contrast be increased. So preprocessing is conducive to the detection of traffic signs.This paper selects two salient features of China road traffic signs namely color and shape to make detection of traffic signs. (1) Proposed a traffic sign detection method based on color segmentation and shape analysis:Firstly, set threshold in HSV color space in order to segment color region, using color block search to position the regions of interest. Color block search narrows search scope and improves detection efficiency. Since the pixels’ color of traffic signs’ graphic border are the same, the method of constructing the edge function is proposed to judge the geometry shape of region of interest. (2) Avoid the weakness of color positioning, here proposes a traffic sign detection method based on shape positioning and color discrimination: First calculate the saturation channel of original image. Using Canny operator on saturation channel for edge detection. Then calculate the shape parameters (degree of circularity, rectangular degree and the extended regular trianglar degree) to judge the shape of edge and locate the position of signs. And then using the modified HSV color space partition model to judge color for classification. Remove non-signs in the classification process. Both the two methods are effectively, but the latter performs better.The basic particle swarm algorithm has drawbacks of easily trap into local optimal solution, the slow convergence in the later convergence phase, worse robustness and so on. The cooperative multiple particle swarm optimization algorithm is proposed to overcome these drawbacks of BPSO. Cooperative multi-PSO can find the feature composition of higher recognition rate more quickly. Therefore, this paper proposes features selection based on the cooperative multi-PSO which with features of Zernike moment and classifier of support vector machine. Zernike moment can better express traffic signs’ characteristics. Use the selected feature composition as input data to train classifier. Then to identify samples with the classifier. The feature selection based on cooperative multi-PSO improves the recognition rate of traffic signs effectively.The experimental results show that the above methods are effectively on traffic sign detection and recognition. The average detection rate reaches 87.5%, and the recognition rate of classifier reaches 91.42%.