結合區塊匹配及Viterbi演算法之影像追蹤系統

Abstract

影像追蹤的應用範圍很廣泛，舉凡機器視覺、自動監控系統、交通流量警報系統等等。對於影像追蹤系統而言，物體運動估測的準確度直接影響整個系統的效能。然而在真實的環境中，若採用區塊匹配演算法來估測物體的運動，其準確度容易受到雜訊的影響。因此在本論文中提出了結合區塊匹配演算法及Viterbi演算法的影像追蹤系統。首先，利用區塊匹配演算法計算影像之間的影光流速度，接下來從影光流速度的分佈中估測馬可夫鏈模組參數，最後利用Viterbi演算法估測影像追蹤系統的物體運動。 經由實驗模擬驗證，結合區塊匹配及Viterbi演算法確實能夠改善影像追蹤系統的抗雜訊能力。

An image tracking system has many applications for machine vision, auto-surveillance system, and traffic flow alarm system. In the image tracking system, the performance depends mostly on the precision of motion estimation. The precision of motion estimation using block matching algorithm, however, is highly sensitive to noise in the real environment. In this thesis, we present a block matching and Viterbi algorithm for image tracking. First, the block matching algorithm is used to compute the optical flow between image frames . Then the model parameters of Markov chain are estimated using the distribution of optical flow. Finally, Viterbi algorithm is employed to estimate the motion by which the image tracking system is developed. Experimental results demonstrate that the noise immunity of the image tracking system is improved.