The Vehicle Detection And Classification In The Video Of Intelligent Traffic Surveillance

The traffic in the 21st century will be intelligent transportation. Intelligent Traffic system (ITS) can rapidly and accurately collects and treats traffic information, makes decision and commands public transportation, making transportation infrastructure to maximize efficiency. Traffic information processing includes vehicle detection and classification, which is an important research direction. After researching related literature from within China and abroad, a simple quickly and accurately algorithm for vehicle detection and classification is proposed in this dissertation. The main contents are as follows:In vehicle detection method, there are four steps: image preprocessing; detection of the region of movement targets; accurate location of movement targets and shadow reduction. Firstly, video image preprocessing was made in order to remove noise; Secondly, the region of movement targets were got by using color images frame difference method, which is different from the traditional method to apply frame different only to two adjacent gray images, after the operation of differential image binarization and morphologic dilation and corrosion algorithm, the more accurate movement target area can be obtained; Thirdly, for the accurate location of movement targets, after researching on the current effect localization algorithms, the vertical and horizontal projection method is used and furthermore improved by locating movement targets with outline areas searching, so the more accurate targets can be obtained; Finally, due to the reasons such as the light, moving targets detection may contain shadow. Because color images provide large amount of information, in this dissertation a new reducing shadow method is proposed by using the same position pixel values in original image to detect shadow in the binarizational image, which can removed the shadow very well and have good usability.The vehicle classification method in this dissertation is to train and test the SVM so as to sort three kind of vehicles: large, mid-size and compact. The input vector of SVM is the features of movement targets obtained from the result of the vehicle detection. These features include the vehicle's length ,width and length-width ratio. The SVM is n classifier which allows using abnormal punishment factor C to classification. The output is three kind of classification.The algorithm can be used in complex scene, such as applies multi-lanes, multi-objective, and can effectively remove interference, satisfy the fast and effective demand of ITS vehicle detection and classification under natural condition. The algorithm has certain theoretical significance and practical value.