MHRGait: Gait Recognition from Momentum Human Rig Pose

Researchers introduced MHRGait, a gait recognition method that uses Momentum Human Rig (MHR) pose estimated from monocular video, with 184 body and hand parameters per frame.

The model organizes controls by anatomy and learns spatial and temporal relationships to represent gait more effectively.

They also proposed MHRGait++, which fuses rig pose and silhouettes using balanced fusion for better accuracy.

Across four benchmarks, the method ranked best among model-based approaches on CCPG and SUSTech1K and transferred well across datasets.

The work suggests compact articulated pose controls can be a low-cost, strong gait representation and a useful complement to silhouette-based systems.