Equine lameness detection is critical for veterinary medicine but faces limitations in subjectivity, cost, and accessibility. While computer vision offers a promising alternative to specialized inertial sensors for extracting movement asymmetries, existing tracking methods either lack the robustness required for equine gait analysis or are computationally prohibitive. We present VFTrack, a novel visual feature tracking algorithm specifically designed for equine lameness assessment. Unlike our previous work which tracked features collectively, this approach tracks each feature independently through a five-stage pipeline—Detect, Match, Prune, Extend, Terminate—integrating ALIKED feature detection with LightGlue matching. This architecture provides robust tracking of anatomical landmarks with O(n \(\cdot \) m) computational complexity. We evaluate VFTrack against three baseline methods (DeepLabCut, CoTracker3, and Omnimotion) on trotting-horse videos. Results demonstrate that VFTrack achieves tracking robustness comparable to the state-of-the-art Omnimotion while operating 120 \(\times \) faster, effectively achieving real-time performance. Compared to DeepLabCut and CoTracker3, VFTrack offers superior tracking duration with minimal computational overhead. By bridging the gap between high-accuracy offline trackers and efficient real-time systems, VFTrack provides a practical, non-invasive, and cost-effective solution for routine veterinary diagnostics. Future work will focus on mobile implementation and expanded validation across diverse equine populations. Our source code is available at https://github.com/kavehsfv/VFTrack_HLD .