Research On Key Technologies Of Detection And Recognition Of Traffic Signs In Real Road

With the development of human society,more and more vehicles on the road have caused many traffic accidents and road pollution problems.In this case,people need an appropriate intelligent transportation system to help them drive.Therefore,as an important part of the intelligent transportation system,the traffic sign recognition system has become a research hotspot.Due to the complex environment in the real road,the traffic sign image acquired by the camera often deforms due to illumination,occlusion,etc.,which will seriously affect the accuracy and real-time of the traffic sign recognition system.Based on the difficulties of the traffic sign recognition system in the real road,this paper studies the key technologies of traffic sign detection and recognition.The specific work is as follows:Aiming at the image preprocessing technology,this paper proposes an adaptive gamma calibration algorithm.This algorithm can adaptively balance the brightness of traffic sign images for different illumination types and enhance the contrast.The experimental results show that this method can effectively improve the color features of traffic sign and reduce the impact caused by uneven illumination.Aiming at the traffic sign detection technology,this paper proposes a combined color segmentation method,which combines the RGB three-component color difference method and the HSV threshold method.The experiment proves that the combined color segmentation proposed in this paper can effectively remove the background noise of the traffic sign image in the real road and segment the traffic sign area.Aiming at the problem that China does not have a complete traffic sign data set so far,this paper has established its own data set of Chinese traffic signs.The data set is expanded by projection transformation.The final data set contains a total of 15703 traffic sign images in 10 different types.Aiming at the traffic sign recognition technology,this paper first analyzes the HOG+SVM based on machine learning and the classic AlexNet network based on deep learning.Based on this,an improved AlexNet network model is proposed.Through comparison experiments,the improved AlexNet network recognition accuracy rate is 98.59%,and the average recognition time of a single traffic sign is 0.0287s.The experimental results show that the improved AlexNe has a significant improvement in training time and recognition time without sacrific-ing recognition accuracy.In addition,this paper also uses the improved AlexNet network to build a traffic sign recognition system.The system is tested by using campus and off-campus road images in different resolution to prove that the system has good robustness and real-time performance in MATLAB environment.