Research & Implement Of Video Vehicle Detection System Based On Surpport Vector Machine

Intelligent Transportation System (ITS) is the necessary approaches for transportation modernization, which are aimed at making full use of available road facilities, improving the reciprocity of vehicles, roads and people, enhancing system security, high-efficiency and cosiness. ITS has proved its high economic and social benefits.Vehicle Recognition System (VRS) is an important part of intelligent transportation system. Nevertheless the application of VRS still maintained in exploration and research stage in China. Due to the different national conditions, direct importation of foreign systems can not satisfy the requirement of complex transport system in China. Under mentioned background, the problems related with vehicle recognition system based on video are researched, and the main work are as follows.A background subtraction algorithm is presented after analyzing and concluding various kinds of vehicle detection algorithms, which can satisfy the dynamic vehicles detection under the static background; To solve the vehicles shadow disturbing effects, the problems existed in current shadow segmentation algorithm are compared, the segmentation based on the genetic algorithm is adopted to segment the image shadow.Using the image registration technology and fusion technology, vehicles image collected by two homogeneity sensors are registered and fused; Based on the principle that gray-scale images from similar sensors have strong relevance, the template matching algorithm is introduced; meanwhile the image pixel level fusion technology is studied, pixel fusion is carried out between vehicle images of two same sensors by using weighted averaging fusion algorithm, which not only meet the requirements of application but makes the operation simple, rapid and easy to realize.For the vehicle classification, edge detection, the morphology filter and link territory processing go into operation, and vehicle figure is obtained after the pre-treatment of the images; vehicle geometry characteristic is calculated according to the relationship between the projection and the geometry mapping; Based on the classification standard, characteristic vector is obtained. Support vector machine theory is applied to video vehicle classification. Because support vector machine is only applicable to binary classification problems, the "one-against-the rest" support vector machine method combined with multiple classifiers binary decision tree method is adopted to build vehicle recognition machine, which solves local extremum problem existed tradition neural network method.These methods proposed in this paper adapt to complex scenes such as large area and multiple objects, and can satisfy the requirement of vehicle detection. They have practicality and certain theoretical meaning, and can be generalized to other fields of video surveillance.