Algorithm Study On Vehicle Foreground Extraction Under Road Traffic Environment

Intelligent transportation systems(ITS) cover all aspects of transportation field. Traffic flow data collection and automatic analysis of traffic behavior based on video images, having been a hot research field, conform to the development of ITS. Vehicle foreground extraction using image processing is considered to be a basic and important step in this hot research field and has important theoretical significance and potential applications in the development of ITS.In this paper, vehicle foreground extraction was divided into two steps: vehicle motion detection and vehicle shadow removal, which have been analyzed and studied for the purpose of vehicle foreground extraction. Firstly, Principle and defects of Gaussian Mixture Model(GMM) were analyzed. In view of the fact that background modeling of GMM is simple and would be affected easily by vehicle foreground, a method of GMM background modeling was put forward based on time-domain constraint average using Grubbs discrimination of outliers. In order to solve the problems of vehicle detection with holes, ghost, and detection errors with vehicle parking,existed in traditional GMM, an algorithm of vehicle motion detection was proposed based on four regions adaptive GMM via improved inter-frame difference and GMM background updating principle. Secondly, it was pointed out that cast shadow of moving vehicle was the shadow studied by this paper after analysis of shadow. In view of the deficiency of shadow removal using color feature solely, regions that certainly belong to vehicle foreground were obtained via difference in Log domain. And the regions of vehicle foreground were used to have constraints on the method of shadow removal based on HSV color space feature. On the basis of these, a method of vehicle shadow removal was proposed based on difference in Log domain and HSV color space feature. Finally, rate of holes filling was defined by this paper based on basic knowledge of mathematics, which was used to analyze quantitatively the degree of solution to the problem of vehicle detection with holes.The experimental results show that the improved GMM algorithm can overcome defects of traditional GMM, the method of vehicle shadow removal proposed by this paper has good performance. It is verified that the full text algorithm of vehicle foreground extraction divided into vehicle motion detection and vehicle shadow removal is reliable.