Research On Multi-region Joint Tracking Human Motion Based On Kalman Filter And Its Application

With the rapid development of computer vision and image processing, object tracking based on video sequences has become hot issues in the field of computer vision. Especially research on human motion has essential influence in both the theory and practical applications. This thesis focuses on tracking human motion in complex background under monocular mobile shooting, and has implemented the speed skating tracking system.As a human motion tracking system, the speed skating tracking system mains includes object feature extraction, motion prediction and matching features. The speed skating tracking system involves numerous discipline domains, such as Computer Vision, Image Processing, Pattern Recognition etc. In this paper, image registration is needed to remove camera movement . The algorithms of object feature extraction and feature matching are key technologies of motion tracking. Advanced feature and feature matching algorithm can improve the precision and accuracy of motion tracking. Meanwhile, motion prediction can reduce the computing time, and improve the robustness of tracking algorithm. Based on discussing popular methods, the following works have mainly been done in this paper, include:1. Image registration is used to eliminate camera movement. Three parts are mainly included in this model: Harris corner detection, feature points matching of the gray related based on multi-constrained conditions, and computing the transform matrix using RANSAC.2. Extract human region feature and feature matching algorithm. By comparison RGB and HSV color space, weighted color histogram is established in HSV color space, and Bhattacharyya algorithm based on color histogram is proposed.3. Establish motion prediction model and occlusion judgment model. The prediction model of Kalman Filter is established based on analysis of the speed skating motion characteristic. Distinct occlusion judgment models and handling occlusion algorithms are proposed to signal-region tracking and multi-region tracking.4. Tracking positioning phase. Mean Shift tracking algorithm and signal-region tracking algorithm based on Kalman Filter prediction are verified, and multi-region joint tracking algorithm is proposed under it. The algorithm takes into account of feature information and motion information of moving targets, which owns an outstanding veracity and tolerance, and high accuracy. Especially, multi-region tracking has higher tracking accuracy on occlusion than signal-region tracking.Experiments show that multi-region joint tracking algorithm based on Kalman Filter in this paper can solve motion human tracking problem in speed skating, and improve the tracking accuracy.