Design And Implementation Of Traffic Flow Tracking Statistics System Based On Video Processing

With the development of social economic,the motor vehicle ownership has risen rapidly,resulting in deteriorating traffic conditions,increasing of congestion and traffic accidents as well as environmental pollution problems.Based on this,research topics such as detection and tracking statistics for road traffic flow have received extensive attention.In this thesis,according to the real condition of traffic,the traffic flow will be analyzed and processed by employing the video image processing technology and deep learning theory.Furthermore,the deep convolutional neural network will be introduced to solve the problem that the traditional algorithm is affected by light changes,multi-vehicles touching and background interference.Also,based on the overlap rate matching and target feature extraction,a vehicle tracking statistical algorithm will be proposed to finally achieve a high-accuracy traffic flow statistical system.The main research works of this thesis are as below:(1)Research on the vehicle detection method based on deep convolutional neural network.Aiming to solve the problems of the high false negative rate and false positive rate in traditional image processing technology due to light and weather changes,multi-vehicles touching,and etc.,the neural network is employed to detect vehicles after a series of SSDnetwork training by the samples of vehicles of different models,appearances and angles from multiple video scenes,which greatly improves the adaptability and accuracy in real scene vehicle target detection.(2)Research on the traffic statistics algorithm based on overlap rate matching.This thesis,after detecting and extracting the moving vehicle,designs a vehicle matching criterion based on the target region boundary overlap ratio,and combines with the virtual coil technology to propose a traffic tracking algorithm.Not only can it solve the counting error problem in congestion section,also achieve a higher algorithm execution efficiency.(3)Research on the relevance judgment method for sports vehicles.In order to solve the problem of the same car repeated counting in the vehicle tracking statistics process due to the inherent non-sustainability of the video frame detection.Speeded Up Robust Features(SURF)point extraction algorithm is employed to extract the characteristics of moving targets as the basis for calculating the similarity of images,and then the vehicle correlation is judged according to the similarity of images.(4)Using SpringMVC as the basic framework for the development of JavaWeb system,the intelligent traffic flow tracking statistics system is designed and achieved based on OpenCV.MySQL serves as a database for storing statistical results.During the development stage of the system,the requirements analysis,outline design,detailed design,system realization and function test were carried out to ensure the stability and practicability of the system.Finally,through the function test and the performance test of the system,it is verified that the traffic flow detection and statistics system has high accuracy and high stability in complex scenarios.The research results can be applied to the actual traffic scenarios and have certain positive implications for the development of transportation industry.