Research on the Technique for Traffic Sign Detection and Recognition
|Course||Circuits and Systems|
|Keywords||Mean shift image segmentation Log-polar transrormDiscrete Fourier transforms Contourlet transform Dual-tree complex wavelet transforn|
With the development of technology and the increased living standards of the people, the number of private car on the highway is growing with each passing day. While the development of transportation has brought great convenience for people, at the same time the problems caused by traffic accidents become more and more serious. The factors that cause the traffic accident include the drivers’error, the weather conditions and so on. According to statistics,80%of the traffic accidents result from the misoperation of driver. In order to reduce the rate of traffic accidents, Traffic Signs Recognition System emerges. The traffic signs recognition system can recognize the traffic sign and then give the driver feedback that allows the driver timely response measures to correct under the changes of the road conditions so as to reduce the traffic accident.The most common types of the traffic signs are warning signs, the directional signs and the prohibition signs. Each type of traffic sign is composed of patterns and characters which have certain colors and shapes. When the illumination intensity or the weather changes, the traffic signs tilt or the collected traffic signs have scale change, they all could affect the recognition effect of the traffic sign recognition system. So the traffic signs recognition system should be of rotation, scale invariance. In addition, the image you grabbed easily contains noise signals, so the system needs to have anti-noise property. In the process of processing image, we use image enhance in the HSI Color Space to reduce the effect of the changes of illumination intensity.Road traffic sign recognition system mainly includes road traffic sign collecting, image preprocessing, image detecting and recognition. By Collecting and recognizing the road traffic signs in different light conditions can verify the stability of the traffic sign recognition system. In order to reduce impacts on the image of road traffic sign from illumination intensity, this paper applied the HSI color space of image enhancement, by the processing of luminance and saturation component so that the image for subsequent processing. In the process road traffic sign, according to the color characteristics of traffic signs, the application of Mean shift image in Lab color space segmentation algorithm, which will be road traffic signs from natural scenes. The combination of log-polar transform and Discrete Fourier Transform (LPT-DFT) has the prosperities of scale and rotation invariance. Contourlet transform and Dual-Tree Complex Wavelet Transform (DT-CWT) possess good multi-resolution and multi-direction, and then they are usually used to extract the texture feature vector. In this paper the method combined by the transformations from LPT-DFT to Contourlet transform and DT-CWT respectively and its applications to Brodatz texture image database verify the advantages, such as scale invariance, rotation invariance and anti-noise property. In addition,the method of combining two kinds of feature extraction based on texture and feature extraction based on color histogram method is used, and then template matching is applied to recognize road traffic sign. The experiments results demonstrate that the methods can effectively recognize the road traffic signs.