Adaptive walk detection algorithm using activity counts
Matin Kheirkhahan, Zhiguo Chen, Duane Benjamin Corbett, Amal Asiri Wanigatunga, Todd Matthew Manini, Sanjay Ranka · 2017
Accelerometers have been the dominant device used for physical activity assessment studies. They are comfortable to wear at different locations and can accurately measure acceleration. Although, accurate methods for detecting walking in the lab and free-living condition using raw acceleration data exist, these algorithms are not useful for determining indoor movements that correspond to short walking bouts (<; 2 minutes). In this paper, we present a new method that is adaptive to a small window of activity count data (10-15 seconds) and robust to within and between subject variability. The adaptive walking detection algorithm is evaluated using 22 adults and walks with a variety of durations ranging from 10 seconds to 8 minutes. The proposed algorithm showed high accuracy for all the walking periods and was significantly better for intervals shorter than 2 minutes.