Activity Recognition And Behavior Pattern Analysis Based On Sensor Data

With the rapid development of Internet of Things technology,pattern analysis based on sensor data provides an effective solution for home security and disease monitoring.Among them,the mode analysis based on ambient sensor has good prospect because of its low cost and low invasiveness.However,the analysis based on ambient sensors has problems including lack of effective sensor sequence segmentation methods and fine-grained analysis of user behavior patterns.Besides,due to intensive deployment and delayed shutdown of sensors,the collected sensor data contains noise,which negatively affects analysis results.In this thesis,from the perspective of signal analysis and pattern analysis,the ambient sensor oriented behavior pattern analysis is investigated based on graph signal processing.The main research work and innovations include:firstly,addressing the lack of effective data denoising methods and the demand for extensive manual manipulation of data pre-segmentation in existing literatures,a sensor sequence analysis method is proposed based on the theory of graph signal processing,which makes effective use of sensor layout information for data denoising,and applies graph wavelet transform to sequence segmentation by modelling sequence information as graph and graph signal.Secondly,in order to enhance the interpretability of the user behavior pattern analysis,a fine-grained behavior pattern analysis and a user classification method is presented based on behavior pattern.A graph comparison algorithm is proposed for user classification.Finally,the validity of the proposed methods is experimentally verified based on the dataset of dementia detection in a smart home scenario.The proposed sequence analysis method reduces the cost of manual data processing.The behavior pattern analysis approach improves interpretability and usability of behavior pattern analysis based on sensor data from the view point of graph.