Development of a Slip Analysis Algorithm: Automating the Maximal Achievable Angle Footwear Slip Resistance Test

Abstract

The use of slip-resistant footwear can prevent falls due to slips, which are a common cause of traumatic injuries. Measuring winter footwear slip resistance with the recently developed Maximum Achievable Angle test currently requires a human observer to identify slips in real time, which is challenging and subject to inter-observer variability. This thesis presents an algorithm for detecting and classifying slips on icy slopes as well as estimating slip distances. A machine learning algorithm was trained and validated using motion capture data from 11,000 steps including 4,700 slips from nine healthy young adults. The overall slip detection accuracy was 91.0%. Slips were classified as one of four types: backward toe slips, forward toe slips, backward heel slips, and forward heel slips with accuracies of 97.3%, 54.7%, 82.6%, and 87.3%, respectively. Finally, the algorithm was able to estimate slip distances with accuracies of 3±5%, 26±67%, 4±6%, and 3±13%, respectively.