The efficiency of sensor fusion techniques depends on the correct knowledge of models describing the processes. The application of these techniques for human motion is far from trivial as there is no straightforward access to the ground truth. We have developed frameworks to deal with this issue based on data analysis, robust statistics, machine learning, and linear Kalman filter.
This scenario becomes even more cumbersome when dealing with motion data from skilled motor performers. These individuals’ biomechanical features and their motion present high diversity, and are usually not reproducible across individuals and activities. Another challenge on improving accuracy on motion analysis is to model and compensate for soft-tissue artifacts.
Applications yet to be published/patented include analyses of baseball pitchers, percussionists, and dancers.
content in Portuguese
invited talk at University of Sao Paulo, department of Mathematics and Statistics: palestra completa em video