ExerSense: Physical Exercise Recognition and Counting Algorithm from Wearables Robust to Positioning

Area of Science:

• Biomedical Engineering
• Sports Science
• Wearable Technology

Background:

• General wearable fitness trackers have limitations in recognizing diverse exercises.
• Previous work introduced ExerSense for real-time exercise segmentation, classification, and counting.
• ExerSense previously supported user-specified exercises with limited motion data.

Purpose of the Study:

• To extend ExerSense capabilities for recognizing multiple physical exercises.
• To evaluate the accuracy of different wearable devices and sensor positions for exercise recognition.
• To determine the optimal device and placement for robust multi-exercise tracking.

Main Methods:

• Collected acceleration data from four different wearable devices during five regular exercises.
• Utilized the ExerSense algorithm for exercise segmentation, classification, and counting.
• Conducted 50 random validations to assess device and position accuracy.

Main Results:

• ExerSense demonstrated robustness across various wearable devices.
• Chest-mounted sensors provided the highest accuracy for the targeted exercises.
• Upper-arm-mounted smartphones, wrist-mounted smartwatches, and ear-mounted sensors showed progressively lower accuracy.

Conclusions:

• ExerSense effectively tracks multiple exercises using wearable sensor data.
• Sensor placement significantly impacts exercise recognition accuracy.
• Chest-mounted sensors are recommended for optimal multi-exercise recognition with ExerSense.