Online Walking Speed Estimation Based on Gait Phase and Kinematic Model for Intelligent Lower-Limb Prosthesis

Intelligent lower-limb prostheses aims to make amputees walk more comfortably and symmetrically which requires the dynamic altering of gait parameters such as walking speed. Some solutions have been proposed such as direct integration and machine learning methods. The former updates walking speed once after an entire gait cycle and the latter collects large amounts of gait data which are unfriendly to lower-limb amputees. Only by using an inertial measurement unit (IMU) placed on the thigh, this paper proposes a novel online walking speed estimation method to determine the walking speed rapidly and accurately in real-time. A step frequency estimator based on the phase variable and a stride estimator based on the inverted pendulum model is designed to determine the walking speed together. The proposed method is evaluated on a public open-source dataset and the gait data were collected in the lab to verify the effectiveness for able-bodied and prosthetic wearers. The experiment results show that the walking speed estimator offers higher accuracy (RMSE of the able-bodied dataset: 0.051 +/- 0.016 m/s, RMSE of the prosthetic wearers' dataset: 0.036 +/- 0.021 m/s) than the previous works with a real-time frequency of 100 Hz. The results also show that the proposed method has good performances both in static speeds and dynamic speed tracking without much data collection before being applied.