| With the rapid development of social economy and the improvement of people’s living standards,the number of motor vehicles is increasing.Urban traffic congestion has become an important factor affecting people’s daily life,and the efficiency of traffic intersections is one of the key factors to solve congestion.One.Traffic intersection traffic light control and intersection guidance line setting is an important research content to solve the traffic efficiency of intersections.The traditional method of traffic science is to use on-site traffic survey methods.With the maturity of current computer methods such as deep learning and the continuous improvement of intersection monitoring video equipment,it is possible to analyze traffic intersection videos based on deep learning methods to obtain traffic intersection flow information.This dissertation uses deep learning methods to study the video analysis model of traffic intersections and develops and implements a traffic intersection video analysis system.The research results of this dissertation include the following work(1)Aiming at the problem of low accuracy of small target recognition of vehicles in video objects at traffic intersections,a vehicle detection model for small targets at traffic intersections is proposed.The model is improved on the basis of the original target detection SSD algorithm,and the deep feature map is up-sampled,and then it is fused with the shallow feature map.In this way,the weighting operation of different feature layers is repeated several times,so that the feature information of small target objects is enhanced,and the accuracy of detecting small target objects is improved.The model was compared and tested on the PASCAL VOC 2012 data set.(2)A small target feature mapping model based on feature enhancement is proposed to further enhance the algorithm’s ability to recognize low-resolution small target vehicles.Through the feature mapping method of small target from low resolution to high resolution,combine GANs with improved SSD network to obtain the feature enhancement algorithm for small targets(GSSD)based on feature mapping,and learn low resolution by generating strong learning ability against network The rate feature is mapped to the feature difference of the high-resolution feature,and the feature obtained by the perceptual network convolution and the feature difference obtained by the generating network are weighted,and the high-resolution feature of the small target object is finally obtained.(3)The realization of a video analysis system for traffic intersections based on deep learning,through the system to achieve the purpose of statistical collection of vehicles in the video.Through the application of traffic intersection video analysis to the demand analysis of traffic flow investigation,the development and realization of a traffic intersection video analysis system based on deep learning.By embedding the above-mentioned algorithm model into the analysis function module,the system realizes the automatic statistical analysis and summary function of different vehicles in all directions of the intersection through the system.The main function modules realized include data collection,data analysis,data summary statistics and system management.And other modules.Figure 37 table 7 reference 78... |
