• gesture recognition
• IMU
• machine learning
• movement tracking
• rehabilitation
• sensors

We were curious to unveil the potential in gesture recognition of a commercial product: an upper body tracking system that was originally meant for gaming. This approach is radically different from what was being done in previous research works, that relied on ad-hoc solutions for motion tracking. We chose four tasks inspired by activities of daily living, which are commonly used to evaluate the recovery of hand function in the field of rehabilitation. With a dataset of 520 hand movements, equally distributed into 4 classes, we were able to reach a classification accuracy of 90%. By means of an interactive platform that provides immediate feedback onscreen it is possible to correctly classify gestures performed by unseen subjects. The main outcomes of this project are:
• The glove could be transformed into a wearable device limited to palm, index and thumb sensors, that would be easier to put on (especially if it is meant to be used by rehabilitation patients) without loss in accuracy.
• The features extracted from the first part of the movement allow to use a shorter acquisition window and therefore to save computational power, while representing a proof of the preshaping strategy adopted by healthy subjects when performing reach and grasp movements.