Investigating the Capitalize Effect of Sensor Position for Training Type Recognition in a Body Weight Training Support System

Load Position Estimation Method for Wearable Devices Based on Difference in Pulse Wave Arrival Time

“…Here, we compare the accuracy of our method with the accuracies of previous studies on estimating the load positions of wearable devices. Vahdatpour et al [ 24 ] obtained an average accuracy of 84% for 10 positions, including the upper arm; Timo et al [ 25 ] obtained an average accuracy of 89% for seven positions, including the left wrist; and Takata et al [ 13 ] obtained an average accuracy of 90% for 10 positions, including the left wrist. In comparison, the accuracy of the proposed method was as high as an F-value of 1.0 for five parts, which is very high.…”

“…A C4.5 decision tree was used to estimate the load positions, and an average accuracy of 94% was achieved. Similarly, Takata et al [ 13 ] had 10 subjects wear a device equipped with accelerometers and angular rate sensors at a total of 10 locations: the head, chest, left wrist, right wrist, waist, front pants pocket, back pants pocket, left ankle, and right ankle. They collected sensor data from the 10 subjects during daily activities such as walking.…”

“…In these conventional studies, a small displacement (e.g., a few centimeters) can be compensated for; however, when the displacement is large or when a sensor is attached to a different limb (i.e., over a joint), the displacement cannot be compensated for, which results in low performance. As for the second category, studies on load position identification classify sensor load positions into predefined body parts [ 13 , 14 ]. However, these approaches require a certain amount of data as input to a classification model, and users must perform specific activities, such as walking for a few minutes.…”