| Daily life activities(ADL)refer to the most basic and common activities that people have to repeat every day in order to maintain their survival and adapt to the living environment.At present,the problem of global population aging is serious.The effective monitoring method of ADL for the elderly can not only be used as an important indicator of nursing evaluation and service in nursing homes or families,but also provide important technical support for the society to deal with the problem of aging.Nowadays,there are many kinds of sensors embedded in smart phones,These sensor data truly reflect the user's life action information.This dissertation studies the method of indoor ADL recognition for the elderly based on smart phones,aiming at only the sensor data collected by smart phones,and fusing sensor data and signals.With the help of processing technology and machine learning technology,the indoor ADL of the elderly can be identified accurately and cheaply.Through the built-in sensors of smartphones,the elderly will produce a lot of useful sensor information in the process of interaction with the living environment.Through the analysis of these sensor information,we can help to obtain the specific behavior of the elderly.In this dissertation,a smartphone-based indoor ADL recognition research framework for the elderly is introduced.A large data distributed processing technology(such as Hadoop)and a service architecture-based back-end ADL large data monitoring platform are integrated to improve the performance and efficiency of ADL recognition.This dissertation studies the location of coarse-grained ADL key activity area KAA(judging whether the elderly are indoors)and the identification of fine-grained ADL in the indoor environment by using different sensor information of smartphones.The efficiency of data storage/analysis and the efficiency of platform architecture are also discussed from the perspective of the efficiency of ADL large data monitoring platform.In order to effectively utilize the sensor information related to the behavior of the elderly to improve the performance of ADL recognition,and to improve the efficiency of ADL recognition by effectively dealing with the bottlenecks affecting the performance of the back-end ADL large data monitoring platform.Specific work includes:(1)Location of coarse-grained ADL key activity areasThe existing research work on Wi-Fi positioning is essentially a static method,that is,many Wi-Fi Access Points(AP)are used as anchors for positioning,and the exact location of each Access Point(AP)is known beforehand.This dissertation presents a multi-scale Wi-Fi localization method based on similarity matching of AP RSSI feature vectors without prior knowledge of AP placement location.A feature vector generation mechanism for target location-oriented detectable AP Received Signal Intensity Indicator(RSSI)is designed.The similarity measure and distance measure are emphatically studied.Based on these two indicators,a new similarity match between the collected AP RSSI feature vectors and the marked AP RSSI feature vectors is realized.The algorithm provides multi-scale location services such as city level,building level and apartment level according to similarity matching results.(2)Fine-grained indoor ADL recognition based on mobile phone light sensorBecause accurate indoor positioning is the key factor to provide indoor ADL recognition related services,this dissertation proposes an indoor ADL recognition method based on smart phone environmental light sensor(LiLo).The purpose of indoor ADL recognition is achieved by analyzing indoor visible light location and frequent pointing of mobile phone.Emphasis is laid on the method of estimating the key active area(KAA)of the elderly based on indoor luminance field map and frequent pointing of mobile phones.On this basis,a smartphone location algorithm based on angle of arrival(AOA)is implemented to obtain the KAA of the elderly more accurately,so that the elderly's specific ADL that associated with KAA can be identified by relying on KAA information.(3)Fine-grained indoor ADL recognition based on mobile phone sound sensorThe occurrence of ADL is usually accompanied by the production of sound,which is the reflection of the interaction between the user and the objects in the environment,and the sound emitted by the objects can also be mapped to the actions taken by the users.Therefore,it has become a feasible research method to recognize users'ADL by using voice recognition technology based on their daily voice(SDL)perception.This dissertation presents an indoor ADL recognition method based on hierarchical perception of sound situation.Based on hierarchical situational perception,audio clips are divided into hierarchical sound situations(AS),sound events(AE)and sound actions(AA).One AS contains several AEs,and one AE contains several atomic A As.This dissertation focuses on the AE/AA fragmentation method,and on this basis,an AE recognition algorithm based on the probability of fragmentation AA is studied and implemented.According to this algorithm,the corresponding SDL of AE is identified,and then the SDL is mapped to the specific ADL to which it belongs.(4)The data storage and analysis efficiency of ADL big data monitoring platformThe ADL big data monitoring platform is based on Hadoop at the level of data storage system.Hadoop data block placement can greatly affect the execution efficiency of Hadoop platform.The existing research work generally adopts block grouping perception to deal with Hadoop data block placement problem.However,most of the existing group-aware placement methods are static and do not take the execution frequency of MapReduce jobs into account.This kind of data block placement method will reduce the potential efficiency of optimal data distribution in the face of MapReduce jobs with high execution frequency,thus affecting performance.From the perspective of how to correlate job execution frequencies,this dissertation proposes a DGAD based on job execution frequencies.Emphasis is laid on the research of job access association model between data blocks based on historical data block access records,and on this basis,a placement method based on association data block clustering is studied and implemented.(5)The architectural efficiency of ADL big data monitoring platformIn order to make the ADL big data monitoring platform have the advantages of flexible configuration,dynamic reconfiguration and lower maintenance difficulty.the platform adopts the architecture of service-based application(SBA),and each function module of the platform is deployed in the form of component services.With the increasing number of platform access users,a single component service(functional module)may not be able to handle more and more access requests,resulting in platform overload and response delay.More and more research work adopts component service replica technology to deal with this problem.With the introduction of component service replica technology,how to place each replica has become an important factor affecting the efficiency of the platform.Most of the traditional replica placement methods are oriented to data replicas.However,component service replicas are more data-dependent than data replicas,that is,there is a lot of communication between computer nodes used to place component service replicas,From the point of view of reducing communication delay by matching component services and computing nodes with similar topological structures,this dissertation proposes a replica placement method of component services based on topological matching.According to similarity clustering method,SBA topological structure acquisition,computing node topological structure acquisition and topological structure matching are carried out respectively.In this dissertation,based on the research of indoor ADL recognition for the elderly,the location of ADL functional area(where ADL occurs)and indoor ADL recognition for the elderly are studied from the perspective of different sensor information of smart phones.In addition,the efficiency of data storage and analysis and the efficiency of platform architecture are also discussed from the perspective of ADL large data monitoring platform efficiency.It can be seen that this study can not only promote the development of ADL identification of the elderly,expand its application scope,and achieve the purpose of improving the quality of life of the elderly,making it possible to slow down the process of social aging from a technical perspective.At the same time,it will make the back-end large data processing platform which adopts similar technologies or frameworks to this dissertation continuously perfect and mature. |
PDF Full Text Request