Research On Motion Vehicle Detection Algorithm Based On The Video Image

With the development of intelligent transportation and information technology,intelligent traffic video monitoring technology has become the main research topic in ITS,and motor vehicle real-time detection is a core part of the subject.The video vehicle detection algorithm can provide the theoretical foundation and understanding for inference and understanding of the traffic behavior and traffic incidents,an effective object detection algorithm is of great significance for the operation of the intelligent transportation monitoring system.In practical surveillance scenario,however,there are many factors influencing the accuracy of the object detection,such as change of light,bad weather and swaying leaves,etc.Therefore,it is necessary to further study an algorithm which the overall detection performance under various scenarios is much better.This paper combines the main problems of current motion vehicle detection,analyzes several advanced background modeling algorithm,finds out the reason of the influence the properties of these detection algorithms,and puts forward an optimal fusion strategy against the residual algorithm and mistakenly identified problems,finally verified by experiments.This paper expounds the present situation of the study on motion object detection algorithm,illustrates the important position of the background modeling.Comparing the detection results of three types of background modeling algorithm,analyzes their respective advantages and disadvantages.This paper proposes a vehicle detection algorithm based on GP.Firstly,this paper expounds the basic concept of genetic programming,characteristics and standard genetic programming method;Secondly,this paper introduces the selection strategy in the process of genetic programming maybe used;Then,mainly elaborates the design process of the detection algorithm based on GP.This paper uses genetic programming to automatically choose the best fusion strategy which includes the image post-processing,logic operation and simple majority voting fusion rules.In particular,various operations contained by the fusion strategy are achieved by a set of unary,binary and n-ary functions which are embedded into the genetic programming framework.This paper gets a specific integration strategy by setting the fusion strategy function parameters,and named GPBF-3(Genetic Programming-based Fusion,n=3)algorithm,then analysis three kinds of tree structure of the fusion strategy.Then let the GPBF-3 algorithm and three types of background modeling algorithm involved in this article make simulation experiments on the six kinds of video sequences in the ChangeDetection.net(2014)CDNET database.Finally,compares with true value calibration images and calculate the evaluation index of the algorithm.The experimental results show that the vehicle detection algorithm based on GP proposed in this paper is effective,and in most scenarios,the new algorithm's robustness and accuracy is better than the rest of the detection algorithm involved in this paper.