Human Motion Activity Analysis Based On Smart-Phone

Human motion activity analysis has always been a hot research direction in the field of information mining. Human motion activity analysis aims at guide people to get a better life experience by human motion data acquisition, data characteristics analysis and mathematical analysis model building. Previous human motion activity analysis work is mainly based on non-contact sensor such as camera. Human motion activity analysis work based on portable contact sensor device has become the mainstream. As a widely spread electronic product, smart-phone has the characteristics of portable, convenient and cheap. Meanwhile smart-phone has the power of large memory, magnificent computing ability and communication capabilities. So it has been used in human motion activity field widely. Main use of smart-phone is to collect human motion data as a platform in this paper. Abnormal state of motion and normal state of motion are concerned. Then, features based on dual tree complex wavelet transform are extracted. Classification algorithm is designed in this thesis. The contents of this thesis are as follows:(1) Data acquisition software application based on android platform and accelerometer, rotation vector sensor embedded in phone is developed. Optimization method is proposed to convert relative coordinate system into absolute coordinate system due to bad interference of relative coordinate system in accuracy of data. The characteristics of data curve under different conditions are analyzed. According to the characteristics of the data during abnormal motion state collected by rotation vector sensor, a detection algorithm based on threshold is proposed. The false alarm rate of the algorithm is extremely low, and it can detect part of abnormal motion state.(2) According to the data of human motion state has characteristics of non-stationary, time-varying displacement sensitivity and periodicity, the dual tree complex wavelet transform is selected as the feature extraction method. The specific implementation scheme, selection of wavelet basis and number of transform layers are discussed. According to the basic theory of dual tree complex wavelet transform, extracted characteristics are analyzed. The feature extraction method can extract the effective information characteristics in the motion state data.(3) According to the characteristics of human motion state, CFS algorithm is chosen for feature selection, which aims at reservation of the most effective features and dimensionality reduction. Compare the effect of the data set of the accelerometer and the rotation vector sensor, and select the best data source. Performance of data set developed by accelerometer and rotation vector sensor is discussed. And the best data source is chosen. According to the characteristics of different classifier, parameter optimization scheme is designed. Under the condition of disappearance of over-fitting, the classifier with the best performance is obtained. Finally, analyzing each classifier according to evaluation principle. Discuss difference between each category of motion. The final classification results accuracy rate is more than 98%.