The Research Of Vehicle Type Based On Video Technology

With the rapid development of economy and urbanization, people’s disposable income continuous rise, the amount of vehicle in China has attained 264 million which ranks second in the world. The continuing rise in vehicle can convenient the allocation of resources and people’s travel, but also brought increasingly serious traffic problems. The phenomenon of road congestion, management confusion and worsening traffic environment has attracted the attention of society and government, seeking a kind of more rapid, efficient and practical way of road management is urgent,so the Intelligent Transportation System(ITS) emerges and achieved rapid development.Vehicle model recognition system is an important part of ITS which provides a large amount of data for traffic management. Compared with traditional radio frequency devices, and other methods utilizing electromagnetic induction, the paper adopted image processing technology based on video which not only easy installation, simple update, and can obtain more information of roads,Through the study of image processing, detection technology, this paper designed a set of automatic identification of models of real-time identification system. The mainly work of paper has the following three parts:(1)Segmentation of the target vehicle:Due to background method to extract the target was fuzzy, frame difference method prone to empty and shape changes, optical flow method’s calculation was complexity and so on. This paper presents an optimized Detection Algorithm for Gaussian model,Build the background model based on time domain and the airspace, then improve the traditional iterative segmentation,select the appropriate threshold to eliminate shadows, avoid situations obtained in the target area is greater than the vehicle.(2) This paper selects texture image entropy, edge density, area, aspect ratio of the image features as features of the vehicle image classification to solve the problems of single feature is difficult to improve the vehicle information expression and distinction.(3)Describes the pattern recognition and support vector machine,and presented a method of optimize the binary vector machine classification, then used it to training the samples, after determining divisibility measure, priority discriminated easily distinguishable categories, and select the appropriate kernel function and penalty factor based on cross-validation.