Research On Traffic Sign Detection Technology Based On Deep Learning

With the development of the economy and the continuous improvement of people’s living standard,automobiles,as an important means of transportation in daily life and production,have been widely popularized,and the number of automobiles in China has been constantly increasing.In order to cope with and alleviate some traffic problems caused by the rapid increase in the number of cars,a series of intelligent traffic systems including Advanced Driving Assistance System,Adaptive Cruise Control System and other advanced application systems have emerged.As an important part of the above system,traffic sign detection is one of the important research directions of computer vision,image processing and machine learning in the field of intelligent traffic.In recent years,due to the popularity of deep learning in the field of computer vision,traffic sign detection and recognition research based on deep learning has attracted more and more attention.In realistic driving scenarios,real-time detection of traffic signs is a challenging task due to the complexity and uncertainty of traffic environment.Therefore,an excellent traffic sign detection system should have high detection accuracy and fast detection speed.In this paper,based on the general target in the deep learning framework for traffic sign detection and recognition problems are studied,the Chinese urban,suburban,high-speed road scene common warning signs,prohibitory signs and mandatory signs the three types of traffic signs as the research object,using Tensorflow deep learning algorithm implementation framework.In this paper,aiming at the detection of small targets of traffic signs,by analyzing the reasons for the poor detection effect of SSD on small objects,the improved model is proposed to use the feature map fusion method to generate a new detection feature map and remove the last two detection feature maps.The experimental results show that the improved model achieves the best performance in traffic sign detection compared with the comparison model and has good robustness.In view of the smooth operation of the model in CPU hardware environment,eight lightweight traffic sign detection models are proposed based on the lightweight convolution model and using different improvement methods.In the experimental environment with only i9 CPU,the eight lightweight model FPS proposed in this paper are all above 10.Finally,the design and development of the traffic sign detection application system are realized.The traffic sign detection model(Tensorflow implementation)is mainly deployed on the mobile experimental platform,and the control interface program is developed by PyQt on the PC.The system mainly integrated model selection,detection results visualization,mobile terminal motion control,model parameter setting,data source selection,results display and other functional modules,which verified the feasibility of the system.