可配合眼球運動之人工眼伺服系統

Abstract

眼球追蹤方面的研究已經發展許多年，除了持續應用在科學研究上，例如討論人眼對視覺刺激的傾向或眼球注視方向反應出的心理狀態等，也積極開發其應用層面，比如做為電腦周邊的輸入裝置或記錄眼睛所注視的影像。市面上目前已發展出多種的眼球追蹤器供實驗研究或各種應用。然而，多數眼球追蹤裝置都有相同的缺點：繁複的校準過程以及準確度容易受頭部轉動影響，以因此至今尚未能夠廣泛應用。 在本論文中，作者設計並實現了一套眼球追蹤裝置系統，能夠將攝影機隨眼球轉動並聚焦在同一物體上，期許未來能夠應用在控制人工眼球的方向以及對焦，此系統使用了一組伺服馬達做為制動器、一組微處理器做為控制器，並選擇一台攝影機做為擷取眼睛影像的輸入裝置，經過影像處理的運算後能夠將系統的兩台分別代表人眼以及未來人工眼球的攝影機聚焦在眼睛所視的同一點上，並經由幾何運算可得到眼睛所視物體相對於人眼之方位以及距離。實驗結果顯示，能夠依據所擷取到的眼球影像成功驅動攝影機轉向人眼注視之物體，又所估測的物體距離與實際距離誤差約在10%以下。

Research in eye tracking technique has been developed for many years. Besides being continuously used in scientific experiments, such as discussing the tendency of human eye for visual stimulation or psychology state revealed from eye gaze direction, it’s application aspects are also being actively developed. Some of its applications include: alternative input devices in human-computer interface or recording of the eye’s gazed object. There are many commercially available developed eye tracking devices for scientific experiments or other applications. However, most of the eye tracking devices have the same defects: complicated calibration procedures and the accuracy is easily affected by head motion. Therefore, it hasn’t been widely used up to now. In this thesis, we have designed and developed an eye tracking system consisting of cameras that follow the eye’s movement and focus on the same target. It is expected to be applied in an artificial eye system controlled by a real eye. The system uses servo motors as actuators, microcontrollers as controllers and a camera to capture the eye’s image. After this image is processed, the two cameras, each represents the real human eye and the artificial eye, are able to focus on a specific point, the same as the human eye’s gazing point. With some geometrical operations, the object’s distance and direction away from human eye is also obtained. As shown in experimental results, the camera representing the artificial eye can be successfully turned to the eye’s gazed object according to the captured eye image and the estimated object’s distance has an error of about 10%.