Level Set Gait Analysis for Synthesis and Reconstruction

We describe a new technique to extract the boundary of a walking subject, with ability to predict movement in missing frames. This paper uses a level sets representation of the training shapes and uses an interpolating cubic spline to model the eigenmodes of implicit shapes. Our contribution is to use a continuous representation of the feature space variation with time. The experimental results demonstrate that this level set-based technique can be used reliably in reconstructing the training shapes, estimating in-between frames to help in synchronizing multiple cameras, compensating for missing training sample frames, and the recognition of subjects based on their gait.