Research On Daily Behavior Recognition Method Based On Smart Phone WiFi-Sound Sensor Fusion Technology

Daily behavioral activities refer to a series of basic activities that people repeat every day in order to maintain personal hygiene and cleanliness,clothing,food,housing and transportation.It is very necessary to monitor the daily behavior of people living alone,especially some elderly people living alone,which can be used as an important indicator for nursing home nursing or home nursing evaluation and services,and can also provide important technical support for society to cope with the aging problem.With the widespread popularity of smartphones and the huge increase in computing power,more and more researchers are trying to use the sensors built into smartphones to monitor and identify human daily activities.Through the built-in sensors of smart phones,users will produce a lot of useful sensor information in the process of interacting with the living environment,and analyzing the information collected by the sensors can help us obtain the specific behavior of the elderly.Due to some problems with the existing technology,this paper studies the WiFi sensor and sound sensor inside the smartphone to locate the user’s location,but how to fully integrate the sound feature and WiFi RSS feature to improve accuracy while controlling the cost still needs to be studied.In this study,a two-stage positioning framework for coarse-grained localization and fine-grained localization of smartphones is used to identify users’ daily behaviors,so as to determine where users’ daily behaviors occur.The specific work is as follows:(1)Locate the subarea where the user resides in a coarse granularityIn the coarse-grained positioning phase,the functional zone where the user resides must be obtained first.In the experiment,by sampling the RSS(signal strength)data information of each wireless AP in different indoor locations,the room was divided into several small grids as the tags of each location,and each RSS data information as the attribute value of the tags,and the two were combined into a set of data.By sampling multiple points,when the new RSS data is obtained,the weighted neighbor algorithm(WKNN)algorithm is used to obtain the label(grid position)information of the current position,so as to achieve the positioning effect.(2)Fine grained daily behavior recognition problem based on mobile phone sound sensorThe occurrence of users’ daily behavior activities is often accompanied by the generation of sound,and the sound emitted by objects can also reflect the user’s behavioral activities.In this experiment,the microphone on the smartphone is used to collect the sound emitted by the user’s behavioral activities,analyze and process the sound,classify the user’s specific behavior,and predict the place where the behavior occurs.First,a clear division of the sound categories that appear in the environment can be made for subsequent classification;Then,a hierarchical situational aware sound classification method is proposed to divide the collected audio data into layers.Sound signals are divided into sound scenarios(AS),sound events(AE),and sound actions(AA).Finally,based on the convolutional recurrent neural network(CRNN)algorithm,the prediction probability is calculated for the type of each AA shard in AE,and the type of AE in the prediction result is the maximum probability type predicted by all AA shards included.By comparing with other acoustic classification algorithms,it is proved that the sound classification algorithm proposed in this chapter can effectively improve the recognition accuracy of daily behavioral sounds,so as to improve the accuracy of daily behavioral activities.(3)Features fusion indoor positioning algorithm based on K-meansIn this article,the positioning and identification of the sound-sensitive sensor and the WiFi sensor inside the smart phone are carried out respectively,and then the K-means algorithm is used to fuse the positioning results of a single WiFi sensor and a single soundsensitive sensor at the decision-making layer,and in the experiment,the positioning algorithm of a single sensor is used as a control,and it is found that the fusion algorithm has higher accuracy and better positioning effect,which fully proves the effectiveness of the fusion algorithm.This article studies how to use the sensor information inside smart phones to locate the location of people’s daily behaviors,and proves the correctness of the method and the effectiveness of the algorithm through experiments,which can provide technical support for human health status recording,disease prevention and patient assistance.