Smart sensors for the recognition of specific human motion disorders in Parkinson's disease

It is proposed a wearable sensing system based on Inertial Measurement Units (IMUs) for the long-time detection of specific human motion disorders. The system uses a single sensor positioned on the head, close to the ear. The system recognizes noticeable gait features as irregular steps and the gait block (freezing of gait). Respect to other positions on the body, the headset has the maximum sensitivity to the trunk oscillations which patients make to get out of the block, increasing dramatically the risk of falls. The headset has also the advantage that it is easy to wear and the whole system can be contained in a single package. In fact, an audio device for auditory feedback to the patient can be integrated without any wireless/wired connection to the ear. The classification of those motion features is performed by an artificial neural network (ANN) and starts from the raw signals collected by the IMU. The ANN algorithm of recognition is extremely versatile and works for any individual gait features. The ANN allows robust and reliable detection of the targeted kinetic features and requires fast and light calculations. In this paper, it is presented the recognition of irregular steps, trunk oscillations and stop state obtained performing calculations out-board on a PC, without losing the generality of the method validity. The final headset system will be extremely energy efficient thanks to its compactness, to the fact that the ANN avoids computational energy wasting, and that the audio feedback does not require any wired/wireless connection. This affects positively the system performance in terms of power consumption and battery life (monitoring time).