Design and Development of a Mobility Recognition System in PD Patients for Tele-Rehabilitation
The purpose of the current work is to design an affordable and accurate device for mobility activity recognition in Parkinson’s patients. The type of sensors and the optimum number were found to minimize the cost while having fair accuracy. 34 activities were selected as the target: LSVT-BIG home training, fast and slow walking, going up and down the stairs, passing obstacles, sit and stand posture, standing up and sitting down, and turning. The collected data was separated within 2.5-second windows. Time and frequency domain as well as wavelet transform parameters were used for feature computation. Through a PCA algorithm, the best combination of features has been extracted. Among 4 classifiers trained for activity recognition, k-NN was the best with the accuracy of 99.7% and sensitivity of 94.1% using left thigh and shank, hip and left forearm sensors.