Enhancing 3D Content Creation in Blender With Monocular RGB Image Keypoints: A Multi-Stage Approach

The standard approach for 3D human models pose animation involved manual frame-by-frame animation techniques. As the industry sought more realistic movements, motion capture emerged as a viable solution, however it requires huge set up and has a lot of constraints. More recently, 3D Human Pose Estimation Deep learning based methods have emerged as a sophisticated solution for generation of 3D human models pose animations from standard inputs. In this paper, our primary objective is to leverage state-of-the-art frameworks for system-atically evaluating distinct components of the pose estimation pipeline and to use the key-findings to support the process of creating a system that can generate motion directly from a single monocular RGB image with a focus on realistic and detailed representation of 3D motion, through a model that can learn from real data showing real and complex human pose. We proposed a four-stage system to generate a 3D Model on Blender using 3D key-points extracted from the input image. The results were promising leading to an efficient pose creation.