The Study Of The Motion Object Detection And Tracking Of The Video Sequence

With the vigorous development of high and new technology, in smart surveillance, film and television manufacture, and many other fields, computer vision has already be an indispensable and important technology. Moving object detection and tracking, as one of the most active and challenging research orientations in this technology, its significance and application value in scientific research and everyday life is becoming increasingly prominent.In this paper, the methods of moving object detection in static background, as well as sigle-target tracking are researched. Through the study of existing research results, in view of the low speed and accuracy in background image extraction, propose a method of building background model by combining RGB space pixel change statistics with median algorithm. For some sequences are difficult to extract background, improve a moving object extraction method by the edge information of frame difference image. According to the target deformation and rotation, be occluded, size change, the interference of light and a number of similar objects, as well as the tracking frame size change, and other issues, proposes a Particle Filter tracking algorithm based on local weighted image. The simulation results show that the two detection methods could extract moving object real-timely, and the tracking method could track the target accurately.1. Propose a background extraction method based on combing RGB space pixel change statistics with time median. Detect the foreground area in three channels(R, G, B) respectively, and then take their and set. Compare the detection result and rate of the proposed method with that of the Gaussian Mixture detection method.2. Improve an object extraction method based on edge information of frame difference image. Us the frame correlation of object motion, combined with edge detection to extract object edge. Extract the foreground region by filling. Compare the two kinds of test result of the proposed algorithm and the related literature algorithm.3. Compare and analyse the three traditional tracking algorithms, they are MS (Mean Shift), KF (Kalman Filter) and PF (Particle Filter). Propose a Particle Filter tracking method based on local weighted image. First, obtain a local weighted image.Ttrack the object uing the weighted histogram of the target area as the feature. Make full use of the local spatial information of the pixel. Simulate and compare the two kind results of the paper method and another tracking method based on Particle Filter algorithm.Experiments prove that background extraction method in the paper can establish accurate background, and object detection results are more accurate than the Gaussian Mixture method, and the detection rate is faster. The detection results using edge information in this paper are ideal, and the extracted object has fine edge. While the target deformation, size change, influenced by strong light and similar things, the tracking method proposed in the paper could also track it accurately. The tracking method can adjust the tracking frame size automatically.