Real-time Detection And Recognition Of Road Traffic Signs Using MSER And Random Forests

The key of the traffic sign detection and recognition system lies in the segmentation of candidate areas of traffic signs,the feature extraction of traffic signs and the design of classifiers.Traffic signs have complex backgrounds and it is difficult to separate them from their backgrounds.Since traffic signs are generally placed in outdoor scenes,they are greatly affected by external reasons(light,damage,shelter,etc.),which greatly increases the difficulty of traffic signs detection.There are many kinds of traffic signs.The same shape of traffic signs varies greatly according to their pictograms,resulting in low accuracy in the identification phase.Although the research on traffic sign recognition systems has begun in the world,and the methods proposed are various,they all have their own shortcomings.The purpose of this paper is to design a real-time traffic sign detection and recognition system.The traffic sign detection and recognition system designed in this paper mainly includes:pretreatment of traffic signs,traffic signs detection and traffic signs recognition.The main work is as follows:(1)For the noise and illumination effects of the traffic sign image acquisition process,the median filter is used for noise reduction,and the gray-scale world method is used to restore the real scene when the image is captured as much as possible to reduce the impact caused by the light.(2)In the traffic sign detection phase.The traditional traffic sign segmentation based on color threshold has the disadvantages of sensitive to light changes.This paper proposes using CLAHE for image enhancement,using MSER to extract the connected components of the image,and obtaining traffic sign candidate regions,according to the geometric features of the candidate region such as area and aspect ratio.The region of interest(ROI)is selected to obtain the traffic sign,and then the Hu invariance moment is used to accurately locate the segmentation and location method.The detection method proposed in this paper can overcome the adverse effects of light on the detection of traffic signs to some extent,achieve real-time requirements,and have higher accuracy.(3)In the recognition stage,this article uses feature fusion to construct the HSV-HOG-LBP descriptor and use random forests to identify traffic signs.Experiments show that the recognition system proposed in this paper has the advantages of good real-time performance and high accuracy.