結合手部辨識的智慧型視力估測系統

Abstract

視力(Visual Acuity (VA)) 量測的主要目的在於檢驗受測者的視覺 清晰度。傳統的視力量測過程中，都會需要一名醫師協助進行測試， 來接收受測者的回應並根據其作答情況進行判斷。在考慮到整個量測 的過程是不複雜且重複性高的情況下，近年來便有許多自動化視力量 測系統出現，希望能以機器取代人力達到節省人力資源的目的。但在 少了人來接收受測者回應的情況下，目前的自動化視力量測系統大多 改採用按鈕或是鍵盤等接觸式設備的方式進行作答。此種非直覺性的 作答方式，很可能使得受測者無法專注於進行視力的量測。再者，接 觸式設備的使用亦會造成衛生方面的問題。為了解決上述改採用自動 化視力量測系統後可能會造成的問題，我們在此篇論文中提出了一個 智慧型視力估測系統(intelligent visual acuity estimation system (iVAE))。 整個系統包含了兩大部分：人眼視力估測演算法以及手部辨識演算法。我們希望藉由感測裝置的使用，配合所設計以揮動速度為參考的手部辨識演算法(velocity based hand motion recognition (V-HMR))，使受測者可以最自然的方式進行作答。而在人眼視力估測演算法的設計上，我們以最大概似估計為核心(visual acuity estimation algorithm using maximum likelihood (VAML))，發展出三種考慮不同概似函數的估測方法。其中，類神經網路(Neural Network) 亦被使用來學習受測者的反應與所看到不同視標大小之間的關係。實測結果顯示，我們所提出的方法可精準且有效率的估測人眼視力。

Visual acuity (VA) measurement is for a subject to test his/her acuteness of vision. Traditional VA measurement includes physician’s assistance, which can be surely replaced by machine since the whole procedure is uncomplicated but repetitious. Therefore, several kinds of automatic VA test are gradually developed and used in recent years. Without experimenter, the traditional way for a subject to speak out or wave a hand in response to the direction of optotype is then replaced mostly by the contact based response such as pushing buttons or keyboards on a device nowadays. However, the contact based response is not intuitive as speaking or waving hands, and it may distract subjects from concentrating on the test. Moreover, the hygienic problem may appear if all subjects operate on the same device. To overcome these problems, we propose an intelligent visual acuity estimation system (iVAE) which keeps the advantage of automatic VA measurement, and also allows subject to respond in an intuitive non-contact way. A velocity based hand motion recognition (V-HMR) algorithm is used to classify hand motion data collected by a sensing device into one of the four directions of optotype. Based on the V-HMR scheme, a visual acuity estimation algorithm using maximum likelihood (VAML) is developed for estimating subject’s vision and is implemented on a tablet. Three sub-schemes of VAML using different likelihood functions is proposed. A supervised machine learning technique, Neural Network, is used for learning human behavior when subject recognizes the optotype. Different response attributes could be considered and implicit characteristic could be found. According to the experimental results, we can conclude that the proposed iVAE system achieve our goals to provide accurate and efficient automatic VA measurements.