Quantifying and Recognising Human Movement Patterns from Monocular Video Images

Quantifying and Recognising Human Movement Patterns from Monocular Video Images

Abstract

Computer vision research into tracking and recognising human movement has so far been mostly limited to gait or frontal posing. This research presents a framework for the spatial and temporal segmentation of continuous motion to quantify and recognise a diverse range of 3D movement skills from gait to saltos. A novel 3D colour body model is accurately sized and texture mapped to each person for more robust tracking. Tracking is further stabilised by estimating the joint angles for the next frame using a forward smoothing Particle Filter with the search space optimised utilising feedback from an Automatic Movement Recognition (AMR) system. A new paradigm enables the temporal segmentation of continuous motion into dynemes for the deconstruction of hundreds of movement skills. Using HMM, the AMR system attempts to infer the human movement skill that could have produced the observed sequence of dynemes. This AMR system, free of markers and set-up procedures, successfully quantifies biomechanical components and recognises hundreds of movement skills. It has also been applied to quantify the Parkinsonian movement disorder and enable biometric identification from gait and a novel anthropometric signature.