PMGS: Reconstruction of Projectile Motion Across Large Spatiotemporal Spans via 3D Gaussian Splatting

The recent study on PMGS presents a groundbreaking approach to reconstructing projectile motion across extensive spatiotemporal spans using 3D Gaussian Splatting. Traditional methods have struggled with complex rigid motion, often limited to short-term deformations. PMGS overcomes these limitations through a two-stage workflow: Target Modeling, which focuses on object-centric reconstruction, and Motion Recovery, which restores motion sequences by learning per-frame poses. A key innovation is the introduction of an acceleration consistency constraint that effectively bridges Newtonian mechanics with pose estimation, enhancing the accuracy of motion recovery. Additionally, the dynamic simulated annealing strategy optimizes learning rates based on motion states, while a Kalman fusion scheme minimizes error accumulation from multiple observations. Experimental results reveal that PMGS significantly outperforms existing dynamic methods in reconstructing high-speed nonlinear rigid motion, un…