Research On Traffic Sign Detection Based On Generative Adversarial Network

With the rapid development of intelligent transportation systems,traffic sign detection has attracted widespread attention in the fields of computer vision and intelligent transportation.In the actual road environment,the detection of longdistance signs faces the disadvantages of small detection target and low resolution,as well as the lack of information caused by bad weather and obstruction.Aiming at the problem of missing traffic sign image information,this paper is based on the improving Faster R-CNN detector,extract the traffic sign region of interest,achieves superresolution of the traffic signs for small targets in the region of interest by combining the target detection algorithm of generative adversarial network(GAN),so as to improve the detection performance and driving safety.Firstly,based on the neural network detection model of candidate region,the paper analyzes the deep learning detection algorithms of R-CNN,Fast R-CNN,Faster R-CNN,select Faster R-CNN detection algorithm as the basis network,analyzes the different feature extraction network structure,the performance of traffic sign detection is tested by different feature extraction networks,determine Faster R-CNN+ Res Net-101 as the traffic sign detection model;The region proposal network(RPN)sets the appropriate number of anchor points for the small target traffic sign size expected to be detected in this paper,the experimentally improved Faster R-CNN detection model has improved detection efficiency compared with the original model.Secondly,in order to solve the problem of small target traffic sign with low resolution and missed or false detection caused by lack of effective information,this paper uses the improved Faster R-CNN detector as the basis and combines GAN as the target detection algorithm to achieve the small target traffic Sign detection,Faster R-CNN detectors are used to generate candidate areas containing small targets,and the fuzzy small targets in the candidate areas are up-sampled in combination with the generative network to generate high-resolution images,the classification loss function and regression loss function are used to improve the discriminative network,which further improves the accuracy of the generated image and increases the detection ability of the target.Finally,experiments on the super-resolution reconstruction performance of small targets based on the improved generative adversarial network in this paper.compared with SRGAN,the experimental results show that the improved generative adversarial network is more clear for small target images.According to the model proposed in this paper and other latest traffic sign detection models,small target traffic sign data sets are tested,the experimental results show that the application of the detection algorithm combining Faster R-CNN with GAN can improve the performance of small target traffic sign detection.