Real-time intention detection of locomotion modes using electromyography

Details

Supervisors

Ronsse, Renaud

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil électromécanicien

Abstract

The good control of a lower limb powered prosthesis is important to allow the amputee to recover a normal and comfortable gait. The high-level control of such a prosthesis can be implemented based on the user's locomotion modes intention. To this end, this thesis investigates the use of electromyography (EMG) to detect the gait intention of healthy subjects. The muscle activities of eight muscles of the thigh were measured and the EMG signals were sampled at a frequency of 1000 [Hz]. Time-domain features were extracted from the signals over time-windows obtained via a sliding window algorithm. The patterns were classified with a linear discriminant analysis (LDA). The best accuracy, 96,43\%, was acquired when four tasks were differentiated at a normal gait cadence: still standing, level-ground walking, stair ascent and descent. Experiments conducted with the same parameters but other subjects scored 92\% and 90\%. However, when adding the ramp ascent and descent tasks, when using a fast or slow gait cadence or when reducing the number of measured muscles, the classification accuracy was significantly reduced. In order for the results to be clinically accepted, a compensation for the EMG's noise sensitivity and non-stationarity is needed. This can be done by adding mechanical sensors as for instance, IMU's or instrumented insoles. Such neuromuscular-mechanical fusion methods are a step further to the good control of powered lower limb prostheses.