Development And Implementation Of Adaptive Activity Recognition System Based On Acceleration Sensor

Human activity recognition technology has been widely used in the fields of identity identification,production control and human-computer interaction.Based on a static model makes traditional activity recognition system rely strongly on the prior knowledge but lack of flexibility and adaptability to suit a particular user.This paper mainly implements an adaptive activity recognition system based on smart phone.According to the references in home and abroad,six daily actions of common human beings were chosen as the actions to be identified in this article,namely,walking,jogging,running,upstairs,downstairs and sitting down.In this study,we propose a simple and adaptive activity recognition algorithm based on the principle of “like dissolves like”.The algorithm treats the three-dimensional(3D)acceleration data flow which is constituted with a serial activity as the material flow with certain molecular structure.Then extract the material's molecular characteristics which can represent different materials,the closer these molecular characteristics values,the more similar the activities.Based on the calculated molecular characteristics values and its confidence level,a reliability-based voting algorithm for human activity recognition was developed.As different activities are different in motion period,a sliding window with variable sizes was created in the algorithm.Furthermore,to adapt to different users,we design an adaptive incremental learning algorithm.The experimental results show that the algorithm has high recognition rate and adaptability.In recent years,the popularity of smart phones is increasing,and its built-in sensor technology is also developing and maturing,which made human activity recognition based on smart phone sensors becoming a research hotspot.So in this paper,we use Android smart phone to develop and implement the user activity recognition system.