Study Of Resident Activity Recognition And Anomaly Detection In Ambient Assistied Living System

With the rapid development of electronic information industry and mobile internet,Ambient Assisted Living system(AAL)is emerging,which involves the collection of sensors,Internet of Things,embedded systems,edge computing,human-computer interaction,offline entity medical and nursing services and other elements.The key technologies of the system,namely the Human Activity Recognition(HAR)modeling Anomaly and Detection,are the core prerequisites for the system to use data to achieve various applications.Focuses on the dense sensing scheme,the author studied the household behaviors,including data source fusion,data feature extraction,modeling and discrimination,and processing methods,respectively,to achieve three temporal granularity levels of human home-level activities discovery,recognition,analysis and anomaly detection.In view of the shortcomings of the original Activities of Daily Livings(ADLs)data structure,such as single attributes,complex time series and manual labeling,a scalable data structure for AAL system is proposed,together with the Super Set Transformation algorithm.We can make full use of the subjective and objective attributes of the existing scheme data sources,and combine external third-party data to enhance the data dimension,thus laying an extended data foundation for subsequent feature selection and extraction.In addition,a superset transformation algorithm is proposed,to map the original sensor and data structure to the new scalable data format seamlessly,which ensure the compatibility and applicability of the system.Taking the deficient of manual label in the field of behavior recognition,the unsupervised method of basic activity discovery is proposed.Combining the advantages of data-driven and knowledge-driven schemes,the 1 to 3 order sensor trigger sequences(distinguished mapping Atomic,Basic and Complex activities)which satisfied the support threshold is mined from the frequent items of sensor sequences triggered in a fixed time segment.After mining and labeling as the black box,the similar activities collected are labeled automatically,which solves the cold-start problem that the classical probability graph model is inavailable without massive manual labeling.According to the three different levels of activities(Atomic,Basic,Complex),we provide specific methods of behavior modeling,discrimination and anomaly detection.Taking the application objectives and scope of application in to account,to realize the feature extraction and time windows in the appropriate pattern.ID3,Hidden Markov Model and Conditional Random Field are used to analyze and model the activity model of different time-granularity offline.Usign the idea of Edge-Computing for reference,threshold and simple discrimination,basic activities dictionary similarity judgment and activity profile similarity descrimination are used to realize the fast response at the front end,which makes up for the shortcomings of the algorithm and realizes the minimum-cost household behavior monitoring and anomaly detection.We implement a case study in the real world to verify the availability and effectiveness of the system via the data result,to verify the hypothesis,model algorithm,discrimination and anomaly detection.This method has been tested and applied in practice.Based on the methods above,combined with the problems of multiple data sources integration and expansion compatibility,a scalable ambient living system is designed and implemented by using hybrid heterogeneous network architecture with minimum total own cost is proposed.Drawing lessons from advanced research experience and taking the objectives and design constraints of current schemes into consideration,a hybrid heterogeneous network scheme which using sensors and configurations of the same type as foreign schemes to achieve basically the same functions,and the price is about one tenth of CASAS schemes.Adopted the cloud edge collaboration mode,the author reduced the total cost through embedded gateway and open source hardwares,replaced the local server with more flexibility,expansibility and effectiveness.