影像之間的差異性比對,在物件追蹤(Objects tracking)的研究領域上,一直是常被探討的重要課題之一,特別是在人體動作的比對上,和一般形狀固定不變的物體比起來更加的複雜,在運算的要求上也更加的嚴謹。SIFT(Scale-invariant feature transform)是目前國內外最常被運用的演算技術,它具有優越的比對能力,足以處理影像之間所發生的各種旋轉、縮放、平移…等在影像比對之中常發生的情況,並具有相當高的穩定性,在本文中,我們使用這項技術來進行研究。 本系統大致上分為兩部分,一是採用SIFT演算法對於物件中的某一特徵點進行追蹤,一是將追蹤完成的軌跡串列作詳細的比對。軌跡的比對方法,是依據”姿勢角度”和”移動向量”作為軌跡資料點的特徵,並以動態時間校正(DTW)找出兩軌跡的最佳比對結果。最後會輸出軌跡比較分析的內容和評分。 In this paper, we propose a tracking method via SIFT algorithm for recording the trajectory of human motion in image sequence. Instead of using a human model that present the human body to analyze motion. Only exact two feature points from the local region of a trunk, one for joints and one for limb. We calculate the similarity between two features of trajectories. The method of computing similarity is based on the “motion vector” and “angle”. We can know the degree of the angle by the connect line from joint to limb in a plane which is using the core of object to be the center. The proposed method consists of two parts. The first is to track the feature points and output the file which record motion trajectory. The second part is to analyze features of trajectory and adopt DTW (Dynamic Time Warping) to calculate the score to show the similarity between two trajectories.