Human Action Unit detection of patient using geometric feature analysis

Human Action Units are based on head, hand and leg movements. Head movements include 11 action units, hand movements show 13 action units whereas leg movements have 16 action units. The measurement parameters are local feature points of body parts and global feature points of whole body. For each body parts such as head, left hand, right hand, left leg, right leg and torso, local descriptors are obtained from the histogram of local features of body parts. Global descriptors includes measurement parameters related to the position and orientation of different parts of body, It also includes area, perimeter, centroid, and eccentricity. Body features are used to detect predetermined action units of head, hand and leg. Rule based logic is developed to detect negative emotion of patient considering 3 head gestures, 4 hand gestures and 4 leg gestures. The proposed algorithm gives promising results for leg action unit detection, moderate results for head and hand action units detection. The body gesture recognition produced promising results. Efficiency of hand movements and leg movements was 90 % where as head movements produced efficiency of 85 %.