Multifactor feature extraction for human movement recognition

In this paper, we systematically examine multifactor approaches to human pose feature extraction and compare their performances in movement recognition. Two multifactor approaches have been used in pose feature extraction, including a deterministic multilinear approach and a probabilistic approach based on multifactor Gaussian process. These two approaches are compared in terms of the degrees of view-invariance, reconstruction capacity, performances in human pose and gesture recognition using real movement datasets. The experimental results show that the deterministic multilinear approach outperforms the probabilistic-based approach in movement recognition.

Research highlights
► We systematically examine multifactor approaches to human pose feature extraction.
► A multilinear approach and a multifactor Gaussian process approach have been adopted and compared.
► We find that the multilinear approach extracts better pose features for movement recognition.