Pose Signal Filtering for Enhanced 3D Gaussian Splatting Animation

This work presents a structured pipeline that leverages Gaussian splatting for precise 3D model reconstruction utilizes the OpenPose library and the ROMP model for motion animation and apply signal smoothing techniques to enhance positional consistency. Those techniques provide the joint motion data movements that drive the animation, often resulting in unstable motion trajectories, causing jitter, noise, and abrupt transi�tions. The proposed approach addresses these challenges by refining the position signal, ensuring smoother and more natural movement. To achieve this, the system applies temporal filtering and motion-aware interpolation to minimize fluctuations in positional data. Using techniques such as low-pass filtering, motion prediction models, and dynamic trajectory adjustments, the pipeline maintains a balance between responsiveness and fluid motion, as mentioned here. This method significantly improves the motion fluidity of animated Gaussian splatting models, making them ideal for real-time applications, digital avatars, and interactive experiences. Although the smoothing process requires additional computational resources, it effectively reduces motion artifacts, resulting in more stable, realistic, and visually seamless animations