Gesture spotting using fuzzy garbage model and user adaptation

Abstract

Recently, many research of HCI have been conducted which enable the elderly and disabled user operate various systems more easily. Operating home appliance by using predefined hand gesture is one example of the research. However, wrong recognition may appear when the predefined command gesture is similar from user’s ordinary behavior. If complex command gesture is adopted to reduce this problem, human friendliness is degraded. For this problem, gesture spotting a task to recognize meaningful gesture from other similar meaningless gestures is considered in this thesis. Fuzzy garbage model is proposed to provide a variable reference value to decide whether the user gesture is the target gesture or not. The reference is achieved from fuzzy target gesture model and fuzzy garbage model which returns the score that shows the degree of user gesture belongs to target gesture and garbage gesture respectively. However, the characteristic of human gesture is different from person to person. Also, the characteristic from single person is changed in different environment. For this reason, user adaptation is required to enhance the recognition capability. In this research, two-stage user adaptation is proposed that off-line(global) adaptation for inter-personal difference and on-line(local) adaptation for intra-difference. For the implementation of the two-stage adaptation, genetic algorithm(GA) and steepest descent method is adopted for each stage. Experiment which recognize left and up command gesture is conducted for 5 different users. The recognition rate of command is over 95% when only one similar gesture exists from command and over 80 % when the command is mixed with many other similar gestures.