Enhancing 3D Human Pose and Shape Estimation in Video Surveillance Scenarios through Improved Alignment with Ground Plane

Existing 3D human pose and shape estimation methods struggle to accurately predict the global 3D position of humans, particularly in video surveillance scenarios with varying camera perspectives and crowded scenes. The proposed RotAvat approach addresses this limitation by aligning the predicted 3D meshes with the ground plane, improving the overall accuracy of 3D human pose and shape estimation in such real-world settings.