Multi Person Pose Estimation and 3D Pose Detection Animation

Pose estimation is the computational determination of an object's exact position and orientation, crucial in computer vision, providing a detailed grasp of its spatial arrangement in 2D or 3D space. In 3D, it involves pinpointing an object's location and pose within a three-dimensional coordinate system, facilitating advanced analysis and real-world interaction. This research paper introduces two novel approaches for pose estimation and 3D animation. The objective is to accurately estimate the poses of multiple individuals captured in videos. The first approach addresses the challenge of simultaneous multi-person pose estimation. To achieve this, we employ the MoveNet Lightning algorithm, which is a cutting-edge deep learning model designed to handle complex pose variations and occlusions. By training the model on diverse datasets, we achieved robust and efficient pose estimation results. This approach enables comprehensive understanding of the poses of multiple individuals in real-world scenarios, opening doors for various applications such as activity recognition, behavior analysis, and interactive virtual experiences. The second approach focuses on generating dynamic 3D animations based on an action captured by two different cameras simultaneously. This setup allows for capturing the action from different viewpoints, which enhances the realism and richness of the resulting animations. For pose detection in each video, we utilize the YOLOv7 algorithm, a well-established object detection framework. By extracting pose information from the frames of both videos, we synchronize the poses and create a coherent 3D animation of the captured action. This approach not only provides realistic visualization of the action, but also facilitates detailed analysis and understanding of the motion dynamics. Through extensive experiments and evaluations, both approaches demonstrate their effectiveness and reliability in estimating poses and generating dynamic 3D animations.