MoPO: Incorporating Motion Prior for Occluded Human Mesh Recovery

Researchers have introduced MoPO, a novel framework designed to enhance human mesh recovery in scenarios involving occlusions. Current methods often struggle with inaccurate poses and motion jitter due to insufficient spatial features for hidden body parts. Inspired by advancements in human motion prediction, MoPO leverages the reliable motion prior inherent in pose sequences. The system comprises two main components: a motion de-occlusion module, which uses a spatial-temporal detector and a lightweight predictor to estimate plausible joint positions for occluded areas based on historical data; and a motion-aware fusion and refinement module, which integrates these completed sequences with image features. Additionally, inverse kinematics are employed to refine the final pose using occlusion-free motion priors. Extensive experiments indicate that MoPO achieves state-of-the-art performance on both occlusion-specific and standard benchmarks, significantly improving accuracy and temporal consistency. This development represents a significant step forward in computer vision, particularly for applications requiring robust human pose estimation in complex, real-world environments where visual data is frequently incomplete or obstructed.