Research On Gesture Recognition Algorithm In The Elderly Care System

The behavior of accidental faint occurs in the elderly with a greater probability. If the elderly cannot get up by themselves and turn to help, it will lead to serious consequences. In addition, the time of the elderly's sleep and exercise is an important indicator of the health. In order to solve the problem of detecting fall behavior and the behavior analysis of the daily life, this paper has studied the related algorithms about feature extraction and gesture recognition and developed the elderly care system.The main process of the system of the elderly care system is: the detection and extraction of the human body, the extraction of the feature vector and the recognition of the human's gesture or action. This paper mainly studies the feature extraction and gesture recognition. Feature is the key to recognize things, in order to classify the human behavior, these characteristics of gesture and action must be extracted. The paper utilized the existing algorithms to extract the simple geometry features, such as the width-to-height ratio and eccentricity, and also extracted the centroid angle feature, the contour moment invariants and the Fourier descriptors with the designed algorithm. The feature vector is composed by the above parameters, which realized the quantization process from image to numerical parameters. The gesture and action recognition is the core part of the elderly care system, this paper uses Support Vector Machine(SVM) algorithm to recognize a gesture based on single frame image, and uses Hidden Markov Model(HMM) with Vector Quantization(VQ) algorithm recognize an action based on multi-frame images. SVM is a linear classifier of two classification, this paper modified the algorithm and applied it to various gusture classification. HMM describes the stochastic process of state transition, the paper modified the VQ algorithm to quantilize the image feature vector, and realized the conversion from feature vector to a speficific number. And with the combination of VQ algorithm, we realized the recognition of continuous actions. If the recognition result is abnormal, a warning is issued. If the result is a regular action, the current gesture or action is recorded and analyzed in a period of time.In this paper, the behavior of the video library provided by Chinese Academy of Sciences and the self-made video library is verified by experiments. It is found that the SVM algorithm is better for the single frame gesture recognition, but it cannot be used for the identification of the action. The HMM algorithm can be of an action category judgment, and recognition results of high-frequency action in life are accurate.