Mutual adaptation in neuro fuzzy system for human posture recognition

Posture recognition is one of the most important techniques for human-robot communication. If robots could read personal intention from the human postures, the communication would become more smooth and natural. Many works of human posture recognition have been published in the literature. Gesture and posture recognition by using Kinect Sensor is an important technique for many types of application. In the related works, many effective techniques composed of feature extraction, classification and pattern recognition have been proposed so far. The processing components are independently designed and evaluated by their own criteria. However, it is difficult to design the components without considering the relationship between each component. Therefore, this paper aims to develop a method of mutual adaptation between information processing components for human posture recognition. In this study, we applied a Neuro-Fuzzy System (NFS) to the classification of human posture. Furthermore, we propose a learning structure that uses Evolution Strategy (ES) to tune the membership function during learning process of the neural network. The proposed method can select suitable membership functions from the candidate solutions generated in ES.