Research On Traffic Sign Detection And Recognition Under Complex Background

Traffic signs are an important road infrastructure,reminding drivers of road conditions ahead and ensuring safe driving.Traffic sign detection and recognition is an important research topic in the field of intelligent driving,providing decision support for assisted driving departments and unmanned driving systems.In complex traffic scenes,small target detection of traffic signs is a major challenge in the field of target detection.In addition,due to light,occlusion,etc.,the image is missing,deformed and blurred,and the recognition accuracy is also low.Faced with these unfavorable factors,this paper based on the Tensor Flow framework,on the basis of Faster R-CNN and capsule network,to achieve the detection and recognition of traffic signs,the specific work is as follows:(1)A traffic sign detection algorithm based on Faster R-CNN is proposed.The core of the Faster R-CNN algorithm is the regional proposal network.The candidate boxes are generated through the anchor points,which greatly reduces the number of candidate boxes and improves the speed of target detection while ensuring accuracy.However,the effect of small target detection is relatively poor.In this paper,the capsule network and Faster R-CNN are merged.First,the shared convolutional layer is improved into a capsule layer,which retains more feature information,and then the anchor points are filtered by feature values,which reduces the number of candidate frames and improves the speed of the RPN network.Finally,the capsule network is used to replace the original classification To improve the classification accuracy.The test results on the GTSDB data set show that the improved Faster R-CNN algorithm m AP is 5.13 percentage points higher than the original algorithm,and the detection speed is increased by 7.13 percentage points.It solves the slower and lower precision of Faster R-CNN traffic sign detection.The problem.(2)An intelligent traffic sign recognition method based on capsule network is proposed.The adaptability of the capsule network itself to image rotation solves the problem of image acquisition angle better.The ultra-deep convolution model is used to improve the feature extraction part of the capsule network structure,and the pooling layer is introduced in the main capsule layer,and the moving index averaging is used.The method improves the dynamic routing algorithm,improves the recognition accuracy of the capsule network in the field of traffic sign recognition,reduces the calculation load,and achieves a superior performance than the original structure.The test results on the GTSRB dataset show that the improved capsule network method’s recognition accuracy in ordinary scenes and complex scenes has increased by 2.36 percentage points and 10.02 percentage points,respectively.Compared with the traditional convolutional neural network,the single image’s The recognition time is shortened by 2.09 ms,which solves the problem of slow recognition speed and low recognition accuracy of the traditional convolutional neural network.