Moving Target Detection And Tracking In The Context Of Static Realization

Dynamic image of moving object offers more information than what single image can provide. Through analysis of many frames of image, we get some especial information which can not get from single image. Base on analysis of dynamic image and combine methods of image detection recognition and tracking, the process of detection-recognition-tracking in moving object detection recognition and recognition became available.This paper is researched for parking system. Under the premise of preprocessing about some defects in original images and taking account some realities of program design, the paper uses Adaptive Gaussian Background Model to simulate moving object's background which is static. With the problems exist in this project, such as the low reaction rate about change of background model and the "ghost" which can not renovate in time or other unexpected results generated by some defects, we not only improve but also simplify the algorithm which is applied in Adaptive Gaussian Background Model to get moving object. These renovation and simplification include: first using some series of video frequency which is applied in the Single Gaussian Background Model to stimulate the initial situation of Multiple Gaussian Background Model. Second give initial definition of a parameter of algorithm. Third adjust the temporal learning rate of weight. The final results prove these solutions not only accelerate the program but also resolve the "ghost" in initial frames of image.By aggregating foreground pixels, the foreground object can be segregate completely, base on which we can gather the characteristics of foreground object respectively. Base on spatial continuity of object, with mathematical morphology and binary connected component analysis, we can wipe off the holes and fill up the noises. Under the premise of complete segregation of object and noise restriction, we apply some characteristics of Single Gaussian Background Model in program design, and then program the proper threshold, owing to these we can get ideal gray background image.Finally, base on area matching and predict algorithm, this paper use methods of borderline tracking and regional growth to predict the object when it is obstructed by other objects. The algorithms, methods and model in this paper are optimized and stimulated, so the system is stable and simple. The final results prove that the operation of these algorithms is faster and the above methods are easy to realize.