The paper introduces the Long-term Effective Any Point Tracking (LEAP) module, which addresses the limitations of existing visual odometry (VO) methods by focusing on dynamic track estimation and temporal probabilistic formulation. LEAP combines visual, inter-track, and temporal cues with anchor-based dynamic track estimation to estimate the reliability of point correspondences and track moving object trajectories in dynamic scenes. The temporal probabilistic approach integrates distribution updates into a learnable iterative refinement module to handle uncertainties and refine point correspondences iteratively. The proposed LEAP-VO system leverages these advancements to handle occlusions and dynamic scenes more robustly. Extensive experiments demonstrate that LEAP-VO significantly outperforms existing baselines across various visual odometry benchmarks, showcasing its effectiveness in complex real-world scenarios.The paper introduces the Long-term Effective Any Point Tracking (LEAP) module, which addresses the limitations of existing visual odometry (VO) methods by focusing on dynamic track estimation and temporal probabilistic formulation. LEAP combines visual, inter-track, and temporal cues with anchor-based dynamic track estimation to estimate the reliability of point correspondences and track moving object trajectories in dynamic scenes. The temporal probabilistic approach integrates distribution updates into a learnable iterative refinement module to handle uncertainties and refine point correspondences iteratively. The proposed LEAP-VO system leverages these advancements to handle occlusions and dynamic scenes more robustly. Extensive experiments demonstrate that LEAP-VO significantly outperforms existing baselines across various visual odometry benchmarks, showcasing its effectiveness in complex real-world scenarios.