Abstract
Modeling complex rigid motion across large spatiotemporal spans remains anunresolved challenge in dynamic reconstruction. Existing paradigms are mainlyconfined to short-term, small-scale deformation and offer limited considerationfor physical consistency. This study proposes PMGS, focusing on reconstructingProjectile Motion via 3D Gaussian Splatting. The workflow comprises two stages:1) Target Modeling: achieving object-centralized reconstruction through dynamicscene decomposition and an improved point density control; 2) Motion Recovery:restoring full motion sequences by learning per-frame SE(3) poses. We introducean acceleration consistency constraint to bridge Newtonian mechanics and poseestimation, and design a dynamic simulated annealing strategy that adaptivelyschedules learning rates based on motion states. Futhermore, we devise a Kalmanfusion scheme to optimize error accumulation from multi-source observations tomitigate disturbances. Experiments show PMGS's superior performance inreconstructing high-speed nonlinear rigid motion compared to mainstream dynamicmethods.