The Recognition Of Traffic Speed Limit Signs In Haze Weather

Most of the traffic accidents are caused by over speed. In recent years there are more and more haze weather in China which brings a significant impact on recognizing the speed limit signs accurately and timely. Therefore, research on the recognition of speed limit signs in haze weather has a vital significance for driving safety.This paper presents an algorithm of recognition of speed limit signs in haze weather, which consists of three parts: dehazing, speed limit signs detection and recognition.Based on the dark channel prior, the proposed dehazing algorithm improves the calculation of the atmospheric light by introducing the largest atmospheric light, which can prevent the calculated atmospheric light to be located into the brighter sky areas and cause more spots and noises after processing. In addition, this paper improves the calculating process of transmission by downsampling with the original image before calculating the transmission map and upsampling to the original size after the calculation of transmission map. This procedure can improve the processing speed on the guarantee of the effect of the original algorithm.For the detection of speed limit signs, this paper segments the red channel in the HSV color space based on the color characteristic of the speed limit signs. Then, morphology processing is used to remove isolated noise. Finally, the circular degree of areas are calculated to find out the candidate coarse positioning of speed limit signs. Furthermore the trained SVM classifier with HOG features is used to find out the interest areas which contain speed limit signs.This paper uses the convolutional neural network to recognize the speed limit signs. The convolutional neural network is one kind of deep neural network which needn’t to extract features manually. In order to recognize the speed limit signs, it uses multiple convolutions and downsampling operations to learn more advanced features from the lower features step by step.This paper simulates the whole algorithm in Matlab platform. Here selects 10000 standard speed limit signs in different scenes to train and test samples. The recognition rate of this paper is up to 98.51%. Experiment results prove that the proposed algorithm can recognize the speed limit signs in different environments, which is robustness and can be used in practical.