Feasibility of Support Vector Machine gesture classification on a wearable embedded device

Abstract

This study looks at the need for a device that monitors upper extremity movements. The device needs to be able to predict gestures on-board to provide immediate feedback. Linear Discriminate Analysis and Support Vector Machine Light and Support Vector Machine Multiclass libraries were implemented on an Atmel embedded system to investigate the limitations. For the Support Vector Machine Light implementation the peak memory usage was 2 Mb and used a sample size of 20 samples per class. The implementation of Support Vector Machine Multiclass with 5 classes, had peak memory usage of 1 Mb and 20 samples per class. The work presented here establishes the ground work for creating a standalone device with on-board classification.