A Speed-Invariant Template-Based Approach for Estimating Running Temporal Parameters Using Inertial Sensors

Segmentation of running data into gait cycles and stance/swing phases is crucial for evaluating running biomechanics. The benefit of magneto-inertial sensors is their ability to capture data in outdoor conditions. However, state-of-the-art inertial-based methods for estimating running temporal parameters are limited to a restricted range of running speeds and, thus, not able to analyze running at very variable speeds. This limitation prevents their use for real-world analysis for a wide range of runners and for sports disciplines where athletes vary their running speed. This study evaluated the speed-dependance of eight relevant foot-mounted inertial-based methods from previous research and proposed a novel method that could be robust to speed changes. The proposed method applied, for the first time, a template-matching algorithm based on dynamic time warping to running analysis and compared it to existing methods. All the implemented methods were tested on 30 runners at different speeds ranging from jogging to sprinting (8 km/h, 10 km/h, 14 km/h, 19-30 km/h) on both treadmill and overground. The most speed-robust performance was achieved by the proposed template-based method, providing estimation errors below 0.1% in stride, between 7%-19% in stance, and between 3%-6% in swing across running speeds. Conversely, all the tested methods from the literature were proved to be significantly speed-dependent. Thus, this study suggested that template-based approach is a valid solution for the inertial-based estimation of temporal parameters during running from slow jogging to fast sprinting. MATLAB codes and templates have been made available online.