Rapid Prototyping of the Learning-based Functionality for Wearable Devices
Ilya B. Gartseev, Ilia V. Safonov · 2017
The variety of wearable devices and the number of their functions have grown significantly over the last 3 years. At the same time the demands of the consumer electronics market require rapid design and production of new devices. In order to speed up the implementation of comprehensive machine learning-based features for such devices and make the software development synchronous with the design of the hardware, we propose a fast prototyping approach for gathering labelled sensor data and implementing a device's functionality. We report a new development platform for wearable devices equipped with inertial MEMS sensors such as accelerometers and gyros. The platform's hardware uses the STEVAL-MKI062V2 board. The software comprises: (1) a Matlab library for interaction with the hardware, (2) software for sensors' data gathering and performing machine learning procedures, (3) software for implementing the functionality of the target device. The benefits of using the prototyping system were demonstrated on the task of recognition of static hand gestures (also called hand postures). In the example, we consider several lightweight classifiers, which can be used for the implementation with hardware that is strongly restricted by computational resources and power requirements.