Research And Implementation Of A Traffic Sign Detection Platform Based On Deep Learning

With the growth of the economy,the total number of cars in China increases year by year,which greatly increases the possibility of traffic accidents and traffic jams.As a new research field,driverless driving is expected to optimize a series of traffic problems brought by cars.The research of unmanned driving is complex and long-term,which includes many aspects,and information collection is one of them.Information collection includes the detection of various key information in the road,including lane lines,traffic lights,traffic signs and so on.This thesis takes traffic sign detection in the field of unmanned driving as the research topic,and focuses on the traffic sign detection algorithm based on YOLO V3,traffic sign detection optimization in complex environment and the realization scheme of traffic sign detection platform.The main contributions of this thesis are as follows:1.Implementation of traffic sign detection based on improved YOLO V3 algorithm.This thesis improves the YOLO V3 network structure by adding feature fusion,inverse feature fusion path and adding jump connection.At the same time,IOU is introduced as the weight to the loss function of the YOLO V3 prediction box to accelerate the convergence of the network.2.Research on traffic sign detection in complex environment.In this thesis,the data set expansion and image preprocessing are carried out for traffic sign detection in fogy days,traffic signs are occluded and dim light.The expansion of the data set includes the sample balancing and sorting of the original data set,the field collection of real special scenes and the image enhancement scheme for the special scenes that are difficult to collect.In the part of image preprocessing,brightness enhancement based on Gamma is adopted for dim scenes,and a determination mechanism based on gray histogram is proposed for whether the image needs brightness enhancement.In addition,for the foggy scene,the equalization histogram is proposed to remove fog.3.Design and implement a Web-based traffic sign detection platform.On the basis of the above research,the traffic sign detection algorithm is applied to the platform.By invoking traffic sign detection algorithm,the platform completes three main functions of online picture detection,video detection and real-time detection.