ActiRecognizer:Design and implementation of a realtime human activity recognition system
In our society,inadequate physical activity is one of severe problems issues for human health,which may increase the health risks of many diseases.Nowadays,smartphones are ubiquitous and widely used around the world,in which multifunctional sensors and wireless interfaces are embedded.Therefore,smartphone is viewed as an appropriate platform for real-time activity recognition to address these healthy problems by monitoring and detecting users everyday activities.In this paper,unlike other wearable devices based applications(e.g.,watches,bands,or clip-on devices),ActiRecognizer,a smartphone-based prototype of a real-time human activity recognition(HAR)is designed and implemented,in which a detailed activity report of individuals(i.e.a pie chart containing the proportion and duration of each activity)can be correspondingly generated based on the detected real-time activities.Specifically,ActiRecognizer adopts client/server(C/S)architecture.At client side,smartphone associated with each individual periodically uploads the accelerometer and gyroscope sensing data to server for activity recognition and monitoring.At serve side,HAR is composed of offline training phase and online activity recognition phase: in training phase,sensing data are collected to extract the desired features that can appropriately characterize behaviors,classification model is generated utilizing these features,and then the trained classification model is used to classify user activity in real time.Finally,detailed activity reports and statistics are available to the user via a secure web interface.
Activity recognition machine learning data mining smartphone
Liang Cao Qun Jin Yufeng Wang Jianhua Ma
Nanjing University of Posts and Telecomm.,China Waseda University,Japan Hosei University,Japan
国际会议
南京
英文
266-271
2017-10-12（万方平台首次上网日期，不代表论文的发表时间）