Spatial-Temporal and Multi-Modal Collaborative Optimization Framework for RGBT Counter-Drone Tracking

This paper tackles the challenges of tracking high-speed maneuvering low-altitude drones amid cross-modal heterogeneity and complex background interference by proposing STMMC, a spatial-temporal and multi-modal collaborative optimization framework for RGBT counter-drone tracking. STMMC integrates three core innovations: the ChronoFlow Estimator (CFE), enabling efficient temporal modeling via self-supervised feature flow estimation; the Inter-modal Feature Refinement Module (IFR), which suppresses modal interference and enhances complementary fusion through dynamic similarity constraints; and the Dynamic Texture-Based Temporal Information Enhanced Training (DT-TIE Training) strategy, boosting robustness to low-saliency targets via physics-constrained motion synthesis. Extensive experiments on Anti-UAV, RGBT234, RGBT210, and GTOT datasets demonstrate STMMC’s superiority, achieving 65.1% precision and 41.5% success rates on Anti-UAV while maintaining real-time performance at 32.2 FPS. Cross-domain tests confirm its adaptability, with 88.4% precision on RGBT234. The framework offers a balanced solution for multimodal tracking in dynamic environments, improving temporal consistency and feature discriminability with an average 2.65%/2.45% PR/SR gain over baselines.