Vision-based hand grasping posture recognition in drinking activity

Drinking activity recognition is not a well-researched area in the human activity recognition area. In this paper, a novel technique to recognize the hand grasping posture in drinking activities is proposed. The proposed method aims to overcome the accuracy issue of Kinect in detecting the correct hand position during drinking activities and no training is required to recognize the grasping posture. Instead, the proposed technique directly extracts the unique features of the grasp posture by using a special Haar-like feature on the input image. By comparing the difference between the total pixel values of each region to a set of thresholds, the grasping posture of the hand can be detected and distinguished from other non-grasping postures or non-hand images. Experimental results indicate that the proposed technique is able to achieve a relatively high accuracy (88% true positive rate and 20% false positive rate) in detecting and recognizing the normal hand grasping posture, which mainly appears in drinking activities where someone is holding a cup.