摘要
眼電訊號 (EOG, Electro-Oculography) 是眾多視線追蹤技術的其中一種，可透過眼球移動所產生的生理訊號來追蹤視線，但其存在著基線飄移 (Baseline Drift) 的問題。基線飄移係生理訊號在測量過程中訊號位準變動，導致眼電訊號偏離眼球轉動量的一種雜訊。目前可用於解決眼電訊號基線飄移的方法相當有限，主要的困難點是眼電訊號與基線飄移的頻譜重疊，難以分離。本論文實現兩種方法來解決基線飄移，第一種是在硬體上利用濾波器電路過濾直流雜訊，然而當眼球轉動並持續地注視某個點時，該濾波器亦會將眼電訊號濾除。第二種不使用硬體上的濾波器來解決基線飄移，而是以高解析度、高動態範圍的類比至數位轉換器 (Analog-to-Digital Convertor) 作為輸入，並且使用小波包轉換 (Wavelet Packet Transform) 來分離基線飄移雜訊。然而，由於眼電訊號與基線飄移雜訊頻譜重疊，故這兩種方法的結果都不甚理想，無法達到穩定的視線追蹤結果，但是可以提供訊號，作為分析視線變化的輸入介面。
並列摘要
Electro-Oculography (EOG) is a kind of gaze tracking method that utilizes bio-signals related to the eyeball rotations to estimate the gaze direction. However, it is well known that the EOG method suffers from the base line drift problem. Baseline drift is a noise uncorrelated with the eyeball rotation. The main difficulty in removing baseline drift is that its spectrum overlays that of the true EOG signal. Therefore, only a few methods are available to alleviate the effect of this problem. In this study, two methods are implemented to reduce the baseline drift noise. The first one is to use a band-pass filter for removing DC components. This will effectively reduce the baseline drift noise with a side-effect that the DC component of the EOG signal will also be removed. Therefore, when a person is staring a stationary point, the band-pass filter attenuates the DC signal and yields incorrect results. The second method is to use sophisticated digital filters rather than the analog band-pass filter. High-resolution and high-dynamic range analog-to-digital convertors are used to acquire the EOG signals. Then, wavelet packet transformation is used to locate and to separate the baseline drift spectrum. However, due to the spectrum overlapping, the second method will still lead to incorrect results. Although both methods are not suitable to provide stable gaze tracking results, the filtered data can still provide important information about how the gaze points variate with the external visual stimuli.