Stream Data Evaluation for a Lifelog Analysis System using a Data Quality Evaluation Framework
暁香 山下, あきか やました, Akika Yamashita · TeaPot (Ochanomizu University) · 2016
In recent years, due to the rapid development of sensor devices, many types of multimedia sensor data can be collected and analyzed to develop useful multimedia applications such as human activity recognition system.Realizing Lifelog which is to memorize life of people as digital data became much easier compared to before.As a result, various Lifelog Analysis Applications using collected sensor data have developed.However, for these Lifelog analysis applications, the quality of input data has not been considered in detail.So, in this study, "Lifelog analysis application that verbalize human action"was actually assumed to be an example, and evaluated how the difference of quality of input data, for example, the frame rate of collected video data or interval of collecting acceleration sensor data or image quality of video data would influence the analysis of result of application.In my experiment, I have dealt with following three kinds of data quality of input data, that is video data and acceleration data, through actual experiment. Data quality A Image quality of each frame of video dataData quality B Obtained number of video data and acceleration data/data quantity per second Data quality C Packet loss rate of video data when communicating through WLAN I have implemented Lifelog Analysis Application and experimented the impact of input data quality on the result of human activity recognition system at real environment -Ocha House.Ocha House is a Japanese home setup to conduct Cyber Physical System experiments using several multimedia camera sensors to monitor and capture human motions and activity data in an end-to-end wireless multimedia network environment.That is, the collected human motions and activity data in Ocha House are transmitted real time over the WLAN to a server for processing and analysis in human activity recognition application.I have evaluated the correlation between input data quality and the result of human activity recognition system Quantitatively.In addition, I have concerned to introduce the Lifelog Analysis Application to each family as abnormal activity detection system inside a smart house.When considering setting up the system in each house, it is effective to collect and analyze the sensor data on Cloud environment.I have also proposed how to implement the system in many houses, and evaluated how to migrate the data quickly and securely.