Research Of Video Motion Tracking And The Application In Augmented Reality

Video motion tracking can automatically detect the target object and follow the accurate tracks of target. It has many applications such as traffic monitoring, video animation, military positioning etc. Augmented reality technology based on video usually requires registering the virtual objects to the target object which should be tracking in time. But it does not widely used on account of the equipments’ high costs. Therefore, the research of video motion tracking and the application in augmented reality have become more important.This paper studies and analyzes the existing target tracking method firstly, then describes the principle and the step of Mean-Shift algorithm in detail. The Mean-Shift can not identify object automatically and the accuracy will drop due to the change of object’s scale. Hence, we propose two improvements. The main research works in this paper are below:1. In order to solve the problem which could not identify object automatically in the first frame of video streaming. A fast AdaBoost target detection algorithm based on optimized weighting parameter is proposed. Firstly, algorithm changes the solving formula of weighting parameter to get low false alarm rates and low false recognition rates. Secondly, dual-threshold is obtained by calculating the feature-value curve. The achieved dual-threshold is used to produce weak classifiers which can form an ensemble classifier. Lastly, we use the ensemble classifier to detect the target in video. Experimental results show that it not only improves the accuracy of detection, but also ameliorates the training and detecting time.2. The issue of scale changing and similarity between target and background will cause low accuracy of Mean-Shift target tracking. This paper improves the method of object tracking using SIFT and Mean-Shift. Instead of SIFT, we use SURF to extract features and weighting to probability density estimate of target model. Then, EM algorithm is used to calculate the Mean-Shift vector for the sake of target tracking. Experimental results show that the proposed method could solve the problems which the change of scale, similarity to the color space, light and rotation effectively.3. We implement the procedure of augmented reality based on target detection and tracking. After the proper detection of target in first frame, we apply image interpolation to magnify or shrink2D models, then2D models can be registered into target which in first frame. Target is be tracked, at the same time use the improved ASM algorithm to locate the contour of target accurately and update the2D models to new location. A procedure of augmented reality is done with above steps.