Movement classification and analysis
The project consisted in developing a proof of concept for the comparison of movement between a reference movement and captured data. It was necessary to demonstrate the hypothesis that it was possible, through the analysis of motion capture data, to classify and analyze movements of physical exercises. To do this, it was necessary to solve problems:
- – In conversion of the movement data that can come from different systems to a common standard;
- – Classification and analysis of movements that can be performed by people of different body sizes, in different orientations, at different rates and over different total durations;
- – In identifying and optimizing tolerance rate metrics, processing times and acceptable error rates.
To solve these problems, we adopted a data-oriented programming approach, based on transform matrix tensors, and allowed robustness to the conditions of motion realization by processing data in terms of the relational characteristics of motion and relational patterns of movement, models derived from computer vision practices. In other words, from complex sequences of motion capture, we have trained intelligent patterns able to detect the beginning and the end of movements like a squat or jumping jack.
Recherche et développement
Programmeurs R&D : Pierre-Olivier Roy , Jean-François Couture and Jérémie Kaltenmark