Improved Method for3D Human Motion Estimation

Despite the challenges posed by noise and occlusion, our new method utilizing a novel conditional VAE model allows for accurate and credible pose sequence reconstruction. Our robust testing approach employs the Model as a powerful motion creator to jointly solve for pose, body shape, and ground/contact. Extensive evaluation demonstrates the method's accuracy and reliability across a variety of inputs, including RGB-D video and 2D/3D joint positions, resulting in impressive motion and contact generation.