Two-layer generative models for estimating unknown gait kinematics

Abstract

We propose a two-layer gait modeling framework for estimating unknown gait kinematics from a monocular camera. Dual gait generative models are introduced to represent a human gait both visually and kinematically via a few latent variables. A new manifold learning method is developed to create two sets of gait manifolds that capture the gait variability among different individuals at both whole and part levels and by which the two generative models can be integrated together for video-based gait estimation. A two-stage statistical inference algorithm is employed for whole-part gait estimation. The proposed algorithm was trained on the CMU Mocap data and tested on the HumanEva data, and the experiments show very promising results on estimating the kinematics of unknown gaits.