Object Detection and Tracking in Video Image
|School||Nanjing University of Posts and Telecommunications|
|Course||Applied Computer Technology|
|Keywords||Target detection Target tracking MeanShift algorithm Centroid kalman filtering|
Moving target detection and tracking image coding techniques as well as one of the core content in computer vision research has important practical value in the field of video surveillance, visual navigation, intelligent transportation, as well as video image compression and transmission, etc.. Moving target detection is the lowest level in the video sequence processing, the the detected effect will directly affect the accuracy of the follow-up of advanced applications. This article describes today's mainstream method of target detection: inter-frame difference method, Gaussian background method. Combined with their respective advantages and disadvantages of the proposed inter-frame difference method based on background subtraction method, and then combined with morphological filtering, and finally extract a moving target. In computer vision, tracking means by calculating the motion sequence image generation of the trajectory of the moving object. There are many ways to target tracking, are mainly calculated object used in the day-to-day application of the most widely used algorithm's undoubtedly MeanShift of continuous change between frames. The method is calculated by iteration The nearest sample distribution of a non-parametric density estimation algorithm. In the majority of cases to ensure accuracy and real-time tracking, is a fast and efficient tracking algorithm. However, the target tracking process mean shift algorithm did not take advantage of the movement of the target direction and speed information, which led unable to accurately track fast target. This paper presents an algorithm based the centroid algorithm MeanShift tracking model. Centroid of the initial position of the moving target and in the centroid position using MeanShift iteration, Israel and Palestine's coefficient to determine the matching degree of the target and the reference target, experimental analysis, the algorithm can be realized quickly and effectively track the target. In addition, when the target motion an obstacle completely blocked this time MeanShift algorithm obstructions mistaken as a candidate target model, again in subsequent frames moving target, MeanShift algorithm and real-time tracking will not again. In this paper, based on the Kalman filter MeanShift algorithm to improve it. Kalman filtering can be provided according to the conventional moving target information to the target initial position of the next moment estimate, then MeanShift algorithm is performed according to the Kalman filter to estimate the initial value of the iteration. By comparison of the improved algorithm with traditional algorithm, the improved algorithm can be completely obscured under the target continuous and stable tracking effect relative to the traditional algorithm has been significantly improved.