An Inertial Multi-Joint Approach for Arm Gesture Recognition

Nowadays, solutions for human gesture recognition are widely diffused and adopted to address specific needs in several application contexts, such as clinical rehabilitation, robotics and Human-Machine Interaction. Among different technologies, including vision systems, ultrasound and infrared sensors, the use of inertial measurement units shows unique advantages, especially related to the ease-of-use and their usability also in not structured environments. This paper deals with the development and the validation of a multi-joint sensing system for arm gesture recognition. The proposed approach exploits commercial sensing nodes, embedding accelerometer, gyroscope and magnetometer, as well as a suitable processing core. Each node provides the quaternions components of its absolute orientation. Such information has been used to feed a Convolutional Neural Network classification model, which aims at classifying ten different gestures. Obtained results demonstrate the suitability of the proposed methodology, showing in the test phase an accuracy equal to 98.33% and a reliability of 99.38%.