Research On Traffic Sign Detection And Recognition Algorithm

As an important part of intelligent transportation system,traffic sign detection and recognition has a broad application prospect in assisted driving and automatic driving.In recent years,with the increasing number of vehicles,road safety and smooth face many challenges,traffic signs detection and recognition become particularly important.How to accurately detect traffic signs is a practical problem facing at present.In this context,this paper improves and optimizes the existing target detection algorithm,and improves the performance of the improved algorithm by combining the unique shape characteristics of traffic signs without increasing the amount of calculation.The main research contents and work include the following two parts:In view of the small scale and variety of traffic signs and the high complexity and poor real-time performance of traditional detection,an improved convolutional network algorithm based on FCOS model was proposed through in-depth study of the FCOS model.In this algorithm,a new feature fusion module is designed,which adds a feature fusion module from low to high on the basis of self-high to low,and adds a cross-link path to increase the degree of feature fusion and improve the quality of small target feature information.In the algorithm detection head,the newly designed fusion module of deformable convolution and channel attention mechanism replaces the original ordinary convolution network,and focuses the feature extraction attention on the discriminant area of traffic sign detection and recognition.Experimental results on traffic sign data set TT100 K show that the improved algorithm has better detection performance and m AP reaches 87.6%.In order to obtain higher quality of traffic signs are sample labels,combining with the characteristics of traffic signs the unique shape,circle,triangle and square,etc.,design a new sample selection strategy,get rid of the noise better tag,low quality and choose more reasonable center of target characteristics of the center,used in the improved algorithm to test the traffic sign recognition.Experimental results show that the new positive sample selection strategy can improve the detection performance of the algorithm and increase the m AP value by 2.1%.