近年來，眼動儀已經是一個常用於心理學分析、疾病分析、廣告配置分析等等領域的一套設備。在本研究中，我們自製了一套穿戴式的眼動儀系統。使用Microsoft HD-6000 相機，改造成可清楚拍攝虹膜與瞳孔邊界的紅外光相機，搭配專用連接器可將攝影機置於眼鏡的鏡框上，讓攝影機能夠拍攝到眼睛的紅外光影像。眼動儀系統最主要的功能是能夠正確的偵測瞳孔的位置，本研究方法使用了以核方法的相關濾波器之物件追蹤演算法實作瞳孔追蹤功能，使用追蹤演算法找出大致上的瞳孔中心位置，並且搭配圓形擬合方法找出正確的瞳孔圓心及半徑，在搭配投影轉換，將瞳孔的位置轉換到實際所看的螢幕位置。在實驗結果中，與手動定位的瞳孔圓心的誤差平均只有2.02 個pixel，半徑大小的誤差為1 個pixel，且在執行速度上，處理一張影像只需要0.0295 秒，相當於每秒可執行33.95 張影像，執行速度超過一般攝影機所能提供的每秒30 張影像，是一套計算快速且準確的眼動儀系統。;In recent years, eye-tracking is already used in areas like psychology, human-computer interface and e-learning. In this study we made a wearable eye-tracking system.We hand-made a IR camera by modifying a commercially available webcam(MS HD-6000) and mounted it to a customized glasses frame. Such device is a wearable eye-tracking system which is able to record a clear video of eye movement when the user wears the glasses frame. The most important feature of an effective eye-tracker is to locate and track the pupil movement correctly in realtime. In this research, we used Kernelized Correlation Filter to implement pupil location tracking. By using KCF and a self-developed circle-fitting algorithm, we are able to detect and track the pupil location accurately. In experimental results,compare manual and automatic detection of the pupil center, the average error of manual and automatic detection pupil center is 2.02 pixel, and the average error ofpupil radius is 1.1 pixel. In terms of execution speed, it only takes 0.0295 second to process an image, which is equivalent to 33.9 FPS (frame per second). Therefore,our eye-tracking system is fast enough to fulfill the real-time requirement and is ready to be used in many practical situations.
