Marionette mass-spring model for 3D gait biometrics

Though interest in gait biometrics continues to increase, there have as yet been few approaches which use modelbased algorithms with temporal 3D data. In this paper we describe a new 3D model-based approach using a marionette and mass-spring model to gait biometrics with 3D voxel gait dataset. To model the articulated human body, we use a stick-figure which emulates the marionettes' motion and joint structure. The stick-figure has 11 nodes representing the human joints of head, torso, and lower legs. Each node is linked with at least one other node by a spring. The voxel data points in the next frame have a role as attractor which able to generate forces for each node and then iteratively warp the model into the data. This process is repeated for successive frames for one gait period. The motion kinematics extracted from this tracking process are projected into the sagittal and the frontal plane and used as a gait feature via the discrete Fourier transform. We use 46 subjects where each subject has 4 sample sequences and report encouraging initial gait classification results.