Human Movement State Recognition Based On Single Acceleration Sensor

Human movement state recognition based on acceleration sensor is a new research in the field of computer pattern recognition.It benefits from the development of micro electro mechanical system and sensor technology and the deep research of pattern recognition theory.Human movement state recognition based on acceleration sensor will play an important role in the fields of motion detection,energy evaluation and health care with the increasing needs in intelligent human-computer interaction and health custody.The essence of human movement state recognition based on acceleration sensor is to have acceleration signals when people exercising with the help of one or more three-axis acceleration sensors,send the data to mobile equipment through Wireless Bluetooth,preprocess the raw acceleration data and extract features of it and finally classify the the motion patterns based on extracting features.However,because of the diversity of objective environment and the complexity of motion,a lot of difficult problems remain to be solved,such as ascertaining the quantities and adorning locations of acceleration sensors,extracting presentative signal features in movement state which is difficult to distinguish and constructing classifiers with high efficiency,accuracy and generalization ability.This thesis has three major focuses in the research of human movement recognition based on acceleration sensor according to above problems:(1)We extract five time-domain features which include mean,standard deviation,Attitude Angle,Time Interval of Peaks and Peak Value and Through Value according to time-domain features and distributing characteristics of acceleration signals and then make a classification recognition of human movement states consisting of staticity,tumbling,running and jumping based on the way of Decision Tree.The result shows that the method of Decision Tree based on five time-domain features has higher accuracy and real-time quality.(2)We propose a way of human movement state recognition based on single fractal and multifractal for problems of distinguishing human ordinary walking,upstairs and downstairs after deep research of non-stationary acceleration signal's structural pattern and self-similar character.We take fractal dimension and generalized dimension as characteristic parameter and achieve the distinction recognition of different movement states with the way of correlation coefficient calculation method.This research proves the effectiveness and feasibility of single fractal and multifractal in distinguishing those three different movement states.(3)The fractal characteristic parameter is extended from single fractal dimension and generalized dimension to fractal matrix through combining fractal theory and wavelet multi-resolution analysis based on multifractal motion state identification.Fractal matrix which is based on wavelet decomposition and reconstruction qualifies component signal's fractal characteristic of states of walking,upstairs and downstairs under different dimensions of mavelet and then explains the complexity and self-similar character of original acceleration signal.The research shows that recognition rate of those three different movements reaches 90% with the premise of less apriori information.