使用平均移動法及卡式濾波器實現即時物體影像追蹤系統

Abstract

在本篇論文中提出一個有效率的物體追蹤系統，在前景和背景分離的前題之下，只紀錄物體的移動資訊．此系統主要分成三個部分，首先是利用平均移動法完成基本物體的追蹤平台，因為相似度比對是建立於特徵空間的統計特性，所以物體的遠近大小形狀變化，並不會影響的追蹤時的準確度，而且只需要移動若干點即可達到區域中最大值，有效降低運算量．為了讓可追蹤物擴大至花色複雜的物體，而且每一張影像的處理時間不能超過1/10秒，以達到即時的要求．因此，我們在系統中加入了平均移動濾波器使影像區塊化，以及利用卡氏濾波器對物體中心的移動做預測．最後採用了四種不同物體來測試，並分析實驗結果．

In this thesis, a real-time object tracking algorithm using mean shift is proposed and implemented. This system mainly consists of three stages. In the beginning, we introduce the computational module based on the mean shift iterations and find the most probable target position in the current frame. The similarity between the target model and the target candidate is evaluated within the statistical data in feature domain. Scale changes and some disturbance of the tracked object are tolerated in this system. To track the objects with a complex pattern, we use the mean-shift filter. The input data will be clustered according to their positions in spatial-feature domain. To take the real-time computation into account, we use Kalman filter to predict the motion vector. Finally, we experiment with four kinds of objects. The results are shown and compared in the end of this thesis.