Detection and Tracking of Moving Object in Video Image Sequences
|School||Beijing Jiaotong University|
|Course||Circuits and Systems|
|Keywords||Target Detection Interframe difference Target tracking Histogram|
The video image processing technology is widely used in industrial, medical, security, management and other fields. Moving target detection and tracking is an important application of image processing technology, image processing of the video sequence, plays an important role in the aerospace, transportation, robot vision, video surveillance in public places passenger data monitoring and other occasions. In this paper, based on the moving target detection and tracking algorithm for the problems of the algorithm is given an improved method of problem-solving, and the effectiveness and stability of the method verified by experiments. Moving target detection is moving target in the current frame image pixels detected, commonly used algorithm for background subtraction, optical flow method, the inter-frame difference method. The inter-frame difference method using the current frame adjacent the continuous moving target regional gray value differences between multi-frame images to detect the target pixel. Traditional two differential detection results of the target area is elongated, deformed, internally generated a large area of ??empty; the three differential detection of moving targets may not clear outline. An odd number of multi-frame difference method presents an improved, for these assays problems, can be relatively accurate shape of the moving target, detect more target contour pixels, and the hollow issues have been improved to some extent. With a detection algorithm for computing successive frames image pixel gray value, by setting the appropriate threshold, the result of the operation is converted into a binary image, and then after morphological image processing, moving object detection results. Accurate moving target detection result is accurate extraction of the target feature the subsequent target tracking process. Moving target tracking is estimated to predict the position of the moving target in the current frame or match recognition. The tracking algorithm is based on the moving target feature extraction based on the target feature information including the location, shape, color, etc.. Tracking algorithm commonly used Kalman filtering, mean shift tracking. Target centroid position of the moving target detection results obtained, and as observations, according to the characteristics of the image scene is constructed state and observation equations of the Kalman filter, the state predicted value as the result of the Kalman filter tracking. Moving target search area hue component in HSV space to build statistical histogram, with the target template matching, Bhattacharyya coefficient maximization conditions, the best matching position mean shift tracking results. Kalman filtering using only the characteristics of the target position, the light and target movement speed of the results of the mean shift tracking, presents an improved moving target tracking method in the analysis of the strengths and weaknesses of the algorithm based on Kalman filter tracking and mean shift tracking results weighted sum of the mean shift algorithm matching template of the target area is constantly updated, the tracking method using the two characteristics of a moving object, the tracking process by adjusting the weighting coefficients to adapt kinds of objective conditions.