ED-FNN: A new deep learning algorithm to detect percentage of the gait cycle for powered prostheses
This publication appears in: Sensors
Authors: V. Thi Thu Huong, F. Gomez Marulanda, P. Cherelle, D. Lefeber, A. Nowé and B. Vanderborght
Number of Pages: 19
Publication Date: Jul. 2018
Throughout the last decade a whole new generation of powered transtibial prosthesis and exoskeletons have been developed. However, these technologies are constraint by a gait detection phase which controls the wearable device in function of the activities of the wearer. Consequently, the gait detection phase is considered to be of great importance as achieving high detection accuracy will produce a more precise, stable and safe rehabilitation device. In this paper we propose a novel gait phase detection algorithm that can predict 100 percent of the gait cycle by implementing an Exponential Delayed Fully Connected Neural Network (ED-FNN) algorithm. Furthermore, due to the forecasting capabilities of the ED-FNN, the system shows no-delayed in detection for real-time applications. The signals fed into the model were taken from only one Initial Measurement Unit (IMU) attached to the lower shank. A large dataset was obtained from 7 healthy subjects performing a daily walking on the flat ground and a 15-degree slope. The results from training individual subjects showed an average of 0.003 MSE (Mean Square Error) in the validation set. MSE values were almost the same both training and validation off-line. When training the dataset from all subjects, the result showed a slight increase in the average with 0.01 MSE obtained from validation set. The method obtained high detection performance compared to the existing algorithms and we propose a concept of gait detection implemented on the prostheses.