Sensor Positioning and Data Acquisition for Activity Recognition using Deep Learning

2018 
In this paper, we perform a study on the sensor positioning and data acquisition details for the HAR system. We develop a framework to support training and evaluation of a deep learning model on human activity data. The activity data is collected in both real-world and lab environments using our testbed system that consists of on-body IMU sensors and an Android mobile device. From the experiment results, we identify that low-frequency (e.g., 10 Hz) activity data is effective for the activity recognition. We verify that four sensors at both sides of wrists, right ankle, and waist can achieve 91.2% recognition accuracy in recognizing ADLs including eating and driving activity. Also, we recognize that two sensors on the left wrist and right ankle are sufficient to present reasonable performance without incurring discomfort in everyday life.
    • Correction
    • Source
    • Cite
    • Save
    • Machine Reading By IdeaReader
    17
    References
    5
    Citations
    NaN
    KQI
    []