全天候之人臉與動作辨識及其於睡著與清醒偵測

Abstract

本論文實現了一套結合人臉辨識、動作辨識與清醒或睡著判別的自動化居家看護系統。首先的人臉與動作辨識工作，待測影像是分別藉由背景相剪法與Haar 疊層分類器產生。為了能抽取出更完整的前景影像，我們分別在灰階與HSV空間建立背景模型。Haar 疊層分類器是一種基於特徵運算的演算法，這種演算法比基於逐點運算的更快速。接著影像將藉由特徵空間與標準空間轉換被投影到一個讓不同類別影像的區別性更大且維度較小的空間。

動作與人臉辨式分別利用模糊法則推論與FisherFace方法來實現。為了將時間軸上的資訊包含進來，我們結合從動作視訊5:1減低抽樣連續三張影像來訓練建立動作辨識模糊法則，並用之推論動作辨識工作。在清醒判別系統中，影像首先會藉由照度隨中心遞減公式來校正。接著利用移動估測方法來量化測試者在睡眠中的活動程度並進一步判定他的清醒/睡著狀態。

In this thesis, we implement an automatic home health care system that combines the face, action and sleep/awake recognition of a person in day and night. The test images are extracted by background subtraction embedded in an action recognition system and then by Haar cascade classifier for face recognition. We build two background models in grayscale and HSV color space to extract the foreground images correctly. Haar cascade classifier for face is a feature-based algorithm that works much faster than the pixel-based algorithm. Then, the test images are transformed to a new space by eigenspace and canonical space projection for better efficiency and separability.

Face and action and recognition is implemented by using FisherFace method and fuzzy rule inference, respectively. We gather three consecutive images 5:1 down-sampled from activity video to construct fuzzy rules inference for containing temporal information to recognize the action. In sleep/awake detection, the LED NIR images will be rectified by using the function of illumination variation firstly. Then, the motion estimation is utilized to quantify the activity degree of a sleeper to determine one’s sleep/awake state.