Recognizing 3D human motions using fuzzy quantile inference

Abstract

Fuzzy Quantile Inference (FQI) is a novel method that builds a simple and efficient connective between probabilistic and fuzzy paradigms and allows the classification of noisy, imprecise and complex motions while using learning samples of suboptimal size. A comparative study focusing on the recognition of multiple stances from 3d motion capture data is conducted. Results show that, when put to the test with a dataset presenting challenges such as real biologically "noisy" data, cross-gait differentials from one individual to another, and relatively high dimensionality (the skeletal representation has 57 degrees of freedom), FQI outperforms sixteen other known time-invariant classifiers. © 2010 Springer-Verlag.