Research And Implementation Of Highway Vehicle Detection And Tracking System Based On Deep Learning

With the development of science,technology and economy,China's highway network has been constantly improved,which brings great convenience to people's life and travel.As the infrastructure of China's modernization drive,the effective operation and management of expressway can ensure the safety of driving and make full use of expressway infrastructure.At present,using artificial intelligence technology to realize ITS is a research hot spot in expressway operation management.Compared with traditional management methods,management methods based on deep learning can achieve higher accuracy,higher real-time performance and more intelligent automated analysis.Since vehicle is the key object of expressway operation management,it is great significant to detect and track vehicles timely and accurately.Therefore,this thesis studies and implements a vehicle detection and tracking system based on deep learning.The main work of this thesis is as follows:1.A data set of highway vehicle detection is constructed including monitoring scenes of expressway trunk lines,toll stations and tunnels,with a total of 31900 pictures.In this private data set,this thesis trains vehicle detectors based on R-FCN and SSD target algorithms respectively,and compares the effect of feature extraction networks of ResNet50 and ResNet101 in the training.Finally,the detector based on R-FCN and ResNet101 for the vehicle detection task in this thesis is determined through comparative experiments.2.The vehicle tracking data set of highway main line is constructed,which contains 9image sequences and 13500 images in total.Referring to the Deep SORT algorithm for multi-object tracking,the SiameseFC algorithm for single-object tracking and the MTCNN algorithm for face detection,In this thesis,a multi-object tracking algorithm suitable for highway application scenarios is proposed,which realizes the tracking task of highway vehicles.3.Based on the research of vehicle detection and tracking tasks mentioned above,this thesis establish a real-time system for vehicle detection and tracking on highways,and realizes the statistics of highway traffic flow on the basis of vehicle detection and tracking.The whole system includes remote service module,background processing module and front-end display module,which can simultaneously detect and track real-time vehicles of four highway surveillance videos.The system separates business processing and data display logically,and visualizes the effect of vehicle detection and tracking,as well as the information of road traffic flow on the front page.