粒子群移動演算法實現高速眼動儀系統

Abstract

眼動儀可以記錄眼球運動、轉換為凝視軌跡，用以研究人類視覺焦點並應用於神經科學、認知心理學、教育、行銷/廣告分析等領域，為現今市場上具有實用性的產品。目前市售的眼動儀多數使用紅外光技術，缺點為環境中的紅外光源會影響系統準確率。因此近年的學術研究投入於可見光眼動儀，但至今市面上尚未出現高精確度的實用性產品。 本系統參考既有的眼球模型，改良眼球模型參數的計算方法與匹配方法以處理一般辦公室光源環境下、480fps 工業用相機錄製的複雜影像。在匹配方法中結合粒子群移動演算法大幅加速計算效率，並改良傳統九點校正使用的二次映射曲線使虹膜中心更加準確地轉換至凝視點，實現 30fps 以上即時處理、平均準確率 1.12 度的系統。

The eye tracking system is a new human machine interface device which can analyze the gaze path by tracking the eyeball movement. It has become increasingly popular in the consumer market, due to the fact that the recorded gaze tracking results can be applied to the study of human attention span, cognitive psychology and in the fields of neuroscience, psychology, education, as well as consumer products. However, gaze tracking system mostly rely on infra-ray (IR) light to enhance the image quality of the eyeball, making the application environments and scenarios of the gaze tracking system an issue to be solved. As a result, more and more researchers devoted themselves to the development of visible light eye tracking system, but very few reliable systems achieve high accuracy. The proposed system aims to improve the system accuracy by modifying the eyeball matching algorithm, even when the images are taken in poor lighting conditions. The improvements come from the modifications of image preprocessing, searching algorithm with the particle swarm optimization (PSO), and a new calibration method. The experimental results indicate the system error is less than 1.12 degree and the entire system reaches processing speed of 30 frames/s.