Research On Reduction And Calculation Method Of Daily Activity Recognition In Smart Home

With the rapid development of the Internet of Things and sensor technology,smart homes have entered people's lives.Smart home usually refers to the living environment equipped with wired,wireless networks and various advanced sensor instruments.In smart homes,sensors are used to obtain data on residents 'physical conditions and daily activities,so as to perceive residents' needs and provide residents with a more convenient and comfortable life.The research on the daily activity recognition of residents based on ubiquitous sensors has injected new vitality into the smart home,making it have more special uses,such as assisting the care of elderly people living alone and life self-care for patients.At present,although research on the activity recognition of residents in smart homes has made some progress,there are still several main following problems:Firstly,in most activity recognition,the accuracy of activity recognition is not high enough,and the accuracy of similar activity recognition is low;Secondly,only the activity recognition of a single user is considered,and the performance recognition of multiple users is not good;Finally,the robustness of the activity recognition model is poor.In view of the above problems,this article conducts the following two aspects of research:(1)In the process of ubiquitous sensor resident daily activity recognition,aiming at the problem of low recognition accuracy of traditional single user activity recognition algorithm under the condition of feature redundancy,this paper proposes an activity feature reduction strategy based on the weight Pearson correlation coefficient(PCC),which selects the sensor features that have a greater impact on the activity for user activity recognition.Firstly,the number of sensor models triggered by activity is extracted as the initial feature set,and the improved PCC is used as the standard of feature reduction,which transforms the features in activity into the correlation between features;then,the reduction formula proposed in this paper is used for feature reduction.This method can calculate the influence weight of the activity characteristics on the activity,and use the characteristics with greater influence on the weight to identify the activity,so the accuracy of recognition is improved obviously.(2)To solve the problems such as low accuracy and weak robustness of existing multi-user activity recognition algorithms,this paper proposes a TF-IDF(term frequency-inverse document frequency)based multi-user activity feature solving method.Firstly,the first triggered sensor model and the last triggered sensor model are recorded as SEF(Start-End-Feature)features of activity recognition;then,record each sensor data in the activity,calculate the weight of the sensor's influence on the activity through the TF-IDF feature solving method,and use the weight of the sensor's influence on the activity as the characteristic of activity recognition.This feature can effectively reduce the influence of falsely triggered sensor on the accuracy of activity recognition.Finally,this paper uses Java and MATLAB to implement the above algorithm,and gets the experimental results in the data set of Weka platform and CASAS of Washington State University,and evaluates and compares the experimental results with relevant evaluation standards,so as to verify the accuracy and accuracy of the algorithm in this paper.