Evaluating real-time methods for pattern recognition of high density electromyography arrays

Our hands are our primary connection to the physical world, but what about the unfortunate people who have lost these amazingly dexterous manipulators? This has spawned an ongoing endeavor to translate biological signals into digital commands to allow natural control of upper limb prosthetics. The persistent challenge in this field is deciphering these biological signals in a rapid enough manner to provide real-time control, while providing accurate predictions, and accomplishing this task with hardware that is both affordable and wearable. The focus of the presented work is the exploration of methods to reduce the computational requirements for real-time electromyography (EMG) based pattern recognition for high density (HD) electrode arrays while increasing the accuracies of predictions. Primarily, this is done using a new feature set that extends time domain (TD) concepts into the spatial domain, while preserving computation simplicity. The proposed methods outperform other advanced feature sets while requiring less than a quarter of the calculations.