Transcript 下載/瀏覽
Posture Recognition with
G-Sensors on Smart Phones
2012 15th International Conference on Network-Based Information Systems
Hui-Huang Hsu , Kang-Chun Tsai
Dept of Computer Science and Information Engineering Tamkang University
Zixue Cheng, Tongjun Huang
School of Computer Science and Engineering University of Aizu
Digital Object Identifier :10.1109/NBiS.2012.135
Date of Conference: 26-28 Sept. 2012 Page(s):588 - 591
Professor: Yih-Ran Sheu
Student : Chan-jung WU
Outline
Abstract
Introduction
Posture Recognition App
Experimental Results and Implementation
Conclusion and Future Work
References
Abstract
Using smart phone to recognize the posture of the user. The app
can record the postures of the user for the whole day and
estimate the burned calories accordingly.
Introduction 1/3
Weight control is a major issue in health management
since overweighting is a very serious social problem in
developed countries
Introduction 2/3
Use the signals from G-sensor in the mobile phone to
identify the postures of the user
Introduction 3/3
System architecture
Posture Recognition App 1/3
Example posture signals
Posture Recognition App 2/3
Artificial Neural Networks(ANN)
sampling period of 0.04seconds
Posture Recognition App 2/3
Artificial Neural Networks
Posture Recognition App 2/3
Hidden note
晴天
陰天
雨天
?
?
腳踏車
?
晴天
?
陰天
?
雨天
?
晴天
?
陰天
?
雨天
?
摩托車
開車
搭車
Posture Recognition App 3/3
Calorie consumption
It is basically the weight (in Kg) of the user times the duration of
the posture state (in hour) and a posture factor
Experimental Results and Implementation 1/3
Experimental Results and Implementation 2/3
The sampling rate is 5 times per seconds. There are
totally 20445 data points in the posture dataset
Experimental Results and Implementation 3/3
The overall classification accuracy is 97 percent
Conclusion and Future Work
The user can be aware of his/her daily activities in a better
way and possibly move more to enjoy a healthier life.
The user’s activity signals are collected and used to train a
personalized neural network model for posture
classification. This should be able to make the classification
accuracy nearly perfect.
REFERENCES
[1]http://www.airitilibrary.com/Publication/alDetailedMesh?docid=16086961
-200812-200907210037-200907210037-286-298
[2] http://developer.android.com/about/index.html
[3] http://developer.android.com/tools/sdk/eclipse-adt.html
[4] http://www.csie.nctu.edu.tw/~kensl/AIrpt.html
[5] http://developer.android.com/guide/components/index.html