Investigating the Impact of IMU Sensor Quantity on Locomotion Recognition Performance Using Neural Networks for Powered Lower-Limb Prostheses.

Powered prostheses provide active assistance, offering greater mobility benefits for individuals with lower-limb loss compared to passive alternatives. A key challenge of the active assistance lies in recognizing locomotion activities accurately with the minimum number of sensors required. Recent studies on neural networks (NNs) have shown significant potential in activity recognition, suggesting the possibility of reducing sensor quantity. This study investigates the impact of inertial measurement unit (IMU) sensor quantity and NN architectures on locomotion recognition using the ENABL3S dataset. The results indicated that, with the same sensor quantity, the HAR-DeepConvLG model achieved the highest F1 score (0.73-0.85) among all the architectures. Besides, the model suggested that accurate recognition of level-walking may enhance the score, as well as increasing the number of IMUs.