Human Activity Recognition Using Intelligent Resonant Accelerometer Employing MEMS Reservoir Computing

This paper reports an experimental demonstration of human activity recognition (HAR) using a MEMS accelerometer integrating a physical reservoir. The doubly clamped beam nonlinear resonator in the frequency modulated accelerometer acted as a physical reservoir. We successfully classified walking and jogging vibration without any software recognition. The physical reservoir resonator holds past input and performs nonlinear conversion for recognition and human activity data that are not linearly separable can be classified by a simple optimization of the output weights by linear regression. The results showed its good performance at the operating frequency where the nonlinear bistability occurs, with the classification accuracies reaching up to 98.8%.