使用單一網路攝影機之視線判斷

Abstract

眼動追蹤一直被應用於認知心理學相關的研究，近年來眼動追蹤更成為人機互動相當熱門的發展重點之一。事實上，眼動追蹤不但能夠被用於輔助行動不便的病患透過電腦與人溝通，也能應用於偵測駕駛精神狀態上，減少駕駛因過度疲勞造成的車禍率，除了可挽救許多人命外，更可降低社會成本。 然而，市面上的眼動追蹤系統經常價格不斐且不易取得，因此我們提出一個只要個人電腦及一個網路攝影機就能使用的眼動追蹤方法。我們修改了Adaboost的人臉追蹤方式，以期調高偵測速度並降低偵測錯誤率，也提出一個能夠快速尋找到虹膜中心位置的方法。最後透過支持向量機，判斷視線可能坐落的區塊，再透過我們設計的視線追蹤機制，進行最終視線所在區塊的判斷。Eye-tracking systems are wildly used in cognitive psychology research problems, and recently eye-tracking technology has been considered as a potential multimedia interaction way. It can be applied to help people who suffer from disease and lost the ability of controlling their movements, so that they can manipulate computers and communicate with others. In addition, eye-tracking system can also be used to detect drivers’ fatigue. Reducing the number of accidents can not only save lives but also decrease society cost. However, commercially available eye-tracking systems usually endure with high cost and hard to fetch problems. We propose an eye movement tracking method using personal computer and single webcam. Our method modifies the Adaboost face detection algorithm to make it faster and reduce the false positive rate. We also provide a new method to calculate the center of iris quickly. Finally, we use SVM to help us categories possible gaze region and determine the final gaze region with our gaze tracking mechanism.