Research On Human Behavior Recognition Based On Acceleration Sensor

In recent years,with the unceasing progress of microelectronics and sensor technology and the deep research in the theory of pattern recognition,human behavior recognition based on acceleration sensor get fast development.With the progress of the society,people demand for intelligent interaction and health care are also growing,so human behavior recognition based on acceleration sensor in areas such as health care,motion detection,energy consumption evaluation has received the widespread attention and research.And corresponding to behavior recognition based on acceleration sensor is that based on computer vision,by contrast,behavior recognition based on the acceleration sensor can reflect the nature of human movement,don't have the limit of scene or time,easy to carry,lower cost and more suitable for popularization and application.At present,in the field of human action recognition,there are generally two classification models:the general model and the personalized model.But the diversity of human bodies is not considered in the general model,so it is not suitable for everyone.On the other hand,the personalized model needs more human intervention.In order to make up the deficiency of the two models,the paper proposes a compromise model training method which trains the raw acceleration data after partitioning the diversified factors of human bodies to obtain multiple models.Additionally,the position of the acceleration sensor has been taken into consideration during the process of data collection,in order to extend the application scope of the recognition model.This method provides the model with better universality and recognition accuracy.Eventually,through the test on the five human actions of standing,walking,running,going up and down the stairs,the recognition rate reaches about 95%.Experiments show that the method is practical and effective.Collect a variety of acceleration data is not a easy thing,so the author studied from algorithm perspective to improve the accuracy of behavior recognition.SVM(support vector machine)was first used to solve the problem of binary classification,it is not very good for multiple classification problem.So this article use the SVM based on binary tree,integrated the SVM and the characteristics of the binary tree,and use the shortest distance algorithm clustering to make the classification more accurate.As a result of the SVM classification precision mainly depends on the selection of kernel function and parameters,so this paper combined the grid search algorithm to choose the punishment coefficient c and kernel parameters y of kernel function.Finally the experimental recognition rate reached more than 90%.Through comparing with the results of other algorithms,it proved that the method is feasible.