A Low-Cost, IMU-Based Real-Time On Device Gesture Recognition Glove

This paper evaluates the possibility of performing fine gesture recognition including finger movements on a low-tech device. In particular, we present a solution with a recognition model that is small enough to fit in the memory of a low- tech device and describe related difficulties associated with this approach. Several different Machine Learning techniques are employed and their individual advantages and drawbacks are explored for the task at hand. Our results indicate an average of 95% accuracy during real-time testing for an eight class decoding task with a custom Recurrent Neural Network approach, that runs on the low-tech device, namely an Arduino Nano 33 BLE. The novelty and strength of this research lies in the fact that we are able to recognize fine hand gestures including finger movements rather than recognizing only coarse hand gestures. The recognition process is conducted on the low-tech device and as a result this solution has all advantages that are typically associated with embedded systems, namely cost-efficiency, battery life efficiency, and a high degree of independence from other devices as well as compatibility with them.