Computer Aided Flying Target Tracking Based on Predictive Dynamic Layer Representations

• Main Authors: Zong-Ying Ho, 何宗穎
• Other Authors: Yung-Nien Sun
• Format: Others
• Language: zh-TW
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/42943338819241965324

碩士 === 國立成功大學 === 資訊工程學系碩博士班 === 94 === 　　In modern defensive techniques, it is an important task to detect and track a flying subject from the acquired image sequence. With the advance in aerospace industry, the speed of flying machine is getting faster. Therefore, the conventional method depended it on manual tracking can not satisfy the demands of on-line tracking. To cope with this problem, we propose a new computer aided tracking method which can track multiple flying targets in an image sequence acquired from an infrared or an optical camera. The proposed method in this thesis can be applied to flying target tracking and many other fields, such as launch and motion analysis, and surveillance system. 　　There are two main topics in the thesis. One is flying target detection, and the other is flying target tracking. However, weather conditions play an important role in flying target detection. The changes in weather condition may influence the complexity in the subsequent image analyses and tracking results. Therefore, it is necessary to evaluate the weather condition in the first step. The weather condition of image scene is defined either clutter or not clutter. According to the weather type, we then utilize different approach to extract the flying targets in the process of target detection. 　　In flying target tracking, we used the dynamic layer to represent the flying target and built the motion, shape and appearance models for each layer. In each iteration, the model parameters were estimated and updated dynamically. In tracking the object, we used the Kalman filter to estimate the central position of object on which the search window of interest is defined. In the experiment of real world image sequences, the proposed method can successfully detect and track the objects.