Research On Algorithm Of Highway Detection System Based On Computer Vision

Intelligent video surveillance is a frontier topic of concern in the field of computer vision, asa new research direction. Without the human intervention, automatic image sequence can beanalyzed as operation of the realization of target motion in dynamic scenes detection, recognition,tracking and behavior understanding. Intelligent video surveillance system overcomes the defectsof traditional monitoring and real-time observation of staff, saving manpower and materialresources, and enhance the accuracy of monitoring judgment. It has broad application prospectsin many important places like military monitoring traffic and banks. And traffic data collectionby using the video technology has become the core content and the key technology in intelligenttraffic system with advantages of large coverage area and rich acquisition of the trafficinformation.The emphasis of the thesis is to extract moving vehicle in the video by detecting andextraction algorithm of moving vehicle, using a classification algorithm based on support vectormachine, by extracting the feature vector of the target vehicle horizontal edge information, todetermine the vehicle models finally. It mainly consists of three parts. In the part of imagepreprocessing, the color images in the video was transformed into gray images, greatly reducingthe computation of image processing in system, and images processed by median filtering. In themoving vehicle extraction part, the comparison between frame-difference method andbackground-difference method was superior. Experimental comparison highlighted theadvantages of background-difference method. In the introduce of background of the key methodof extraction algorithm, the average background modeling algorithm, compared by other twoalgorithm, was more suitable for real-time traffic detection system. And by comparing thedetailed description of the algorithm in the concrete realization, it need to pay attention to thedetails of the process in the algorithm. In models determining part, firstly the traditional edgedetection algorithms were compared together, then a fast edge extraction algorithm suitable forreal-time traffic detection was presented, illustrating its advantages and feasibility by experiment.Then a classification algorithm based on support vector machine was proposed, which was basedon the fast edge extraction and level edge algorithm in theory and its feature vector wasregularized by linear interpolation. Experiments verified the feasibility and effectiveness of theclassification algorithm.The proposed algorithm was based on support vector machine models and linear classification method and the feature vectors of it were extracted from the horizontal edgeinformation. It is suitable for real-time traffic detection system for the little cost and beingfeasible and effective. Although this study was only on two categories of small passengervehicles and large-and-medium-sized vehicles, the algorithm can be applied in a smaller range ofvehicle detection and decision system.