特徵辨識與匹配運算掌握了大部分即時立體視覺系統的效率，如何正確地使用極限幾何求得共軛對成為了匹配運算法優良與否的關鍵，但是假若拍攝影像的深度與範圍差異不是很大，例如手術房應用或平台量測，這類小範圍內的眾多匹配點的拍攝影像將會導致系統效率不彰。本研究透過紅外線攝影方式簡化特徵辨識的複雜度，並針對有限深度區域提出了影像匹配方法的改良，保留了影像扭正的優點，利用影像套合技術提升傳統扭正的效率以進行快速的匹配工作，根據實驗結果，本研究所提出的改良方法可成功地降低匹配候選點的數量，約可將匹配準確度提升 50%，效率亦可增進 100%。

Both feature detection and point correspondence algorithms always heavily influence the performance of real-time tracking in general binocular systems. In this paper, we develop an infrared tracking system to reduce the deficiencies of feature detection. When this system is implemented in the operating room, light reflection from the metal instruments causes computational redundancy during point correspondence. General epipolar rectification with image registration techniques improves the efficiency of correspondence dramatically. The total amount of candidate correspondences can effectively be reduced by using the image registration method for point correspondence. Our algorithm takes the advantages of both rectification and image registration to achieve real-time tracking. Moreover, the proposed method improves accuracy by approximately 50% and efficiency by 100%, in comparison to general epipolar rectification.