Research On Smartphone Sensor Data Based Activity Recognition

In the context of intelligent life,the research of different activities recognition based on the data of smartphones' built-in sensors has attracted extensive attention from researchers in recent years.Although these studies have a good recognition effect on the distinction of short-term actions,they often fail to achieve good performance while perceiving continuous activities.And this kind of work generally exists the disadvantages of small data mining intensity and low data utilization rate.Based on the survey of a great deal of work related to activity recognition,this thesis analyzes the feasibility of using built-in sensors to perceive continuous activities.Through designing an effective data acquisition system and combining specific practical problems,this work has been thoroughly studied.The main contributions of this thesis mainly include:(1)We analyze and discuss the general requirements of this kind of work on sensor data and the method of data collection.By using the results of the analysis,we implement a sensor data acquisition system based on the android platform.Also,extensive tests have been carried out on different types of equipment.The results show that the data collection method designed in this paper can achieve the purpose of stable and effective data collection.And the data collection system also has the characteristics of convenient operation and low intrusion.(2)Using the developed data collection platform,this paper makes an in-depth study on the problem of continuous posture monitoring based on the research summary of the general activity identification framework and designs a new continuous state perception system.The extensive experimental results on the real data set show that the system designed in this thesis can accurately perceive the posture of using a smartphone when lying down.(3)Taking the actual map contour generation as the research object,this paper deeply analyzes the problem of low data utilization that generally exists in the work of activity recognition.A data fusion method is proposed,and the performance test is carried out on the data collected from real road sections.The test results not only verify the effectiveness of the series of data processing algorithms and activity perception models designed for generating map contour sequences but also prove the feasibility of deeply mining the sensor data generated by the continuous activity perception process.