A Method for Real-Time Prediction of Gait Events Based on Walking Speed Estimation

The acquisition of human gait information is pivotal for controlling lower limb exoskeletons. Despite the current richness in gait event detection methods, the associated detection algorithms inherently exhibit unavoidable latency. Therefore, this study introduces a real-time approach for predicting gait events, specifically initial contact (IC) and toe-off (TO), leveraging a novel algorithm for real-time estimation of walking speed. In the offline phase, a gait event prediction regression model is derived from walking experiments conducted under discrete velocity conditions. In the online phase, an electromyography (EMG) sensor and an inertial sensor are utilized to precisely identify the early feature points of gait events, while walking speed estimation is realized using a single inertial sensor, ultimately combining with the prediction model to anticipate gait events under varying walking speeds. Real-time experimental results from five subjects showed that the recognition rates for the early feature points of the IC and TO events reached 100% and 99.2%, respectively. For gait event prediction, the mean prediction errors (PEs) are 4.6 ms for IC and 2.9 ms for TO, with a mean velocity estimation error of 0.003 m/s. Accurate gait event prediction supports applications such as exoskeleton motion control and precise mode transitions.