Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MATrack: Efficient Multiscale Adaptive Tracker for Real-Time Nighttime UAV Operations

Xuzhao Li, Xuchen Li, Shiyu Hu

TL;DR

The paper tackles the critical problem of reliable real-time nighttime UAV tracking, where low light, clutter, and dynamic viewpoints degrade performance. It introduces MATrack, a multiscale adaptive tracker built from three synergistic modules: Multiscale Hierarchy Blender for robust cross-scale fusion, Adaptive Key Token Gate for dynamic emphasis of object signals, and Nighttime Template Calibrator for stable long-term template updates, all operating on an Overlapped ViT backbone. MATrack achieves state-of-the-art results across multiple nighttime benchmarks, including UAVDark135, NAT2024-1, NAT2021, and DarkTrack2021, while maintaining real-time speed at 81 FPS. Real-world UAV tests confirm its robustness and practicality for missions such as nighttime search and rescue and border patrol, highlighting its potential for deployment in resource-constrained robotic systems.

Abstract

Nighttime UAV tracking faces significant challenges in real-world robotics operations. Low-light conditions not only limit visual perception capabilities, but cluttered backgrounds and frequent viewpoint changes also cause existing trackers to drift or fail during deployment. To address these difficulties, researchers have proposed solutions based on low-light enhancement and domain adaptation. However, these methods still have notable shortcomings in actual UAV systems: low-light enhancement often introduces visual artifacts, domain adaptation methods are computationally expensive and existing lightweight designs struggle to fully leverage dynamic object information. Based on an in-depth analysis of these key issues, we propose MATrack-a multiscale adaptive system designed specifically for nighttime UAV tracking. MATrack tackles the main technical challenges of nighttime tracking through the collaborative work of three core modules: Multiscale Hierarchy Blende (MHB) enhances feature consistency between static and dynamic templates. Adaptive Key Token Gate accurately identifies object information within complex backgrounds. Nighttime Template Calibrator (NTC) ensures stable tracking performance over long sequences. Extensive experiments show that MATrack achieves a significant performance improvement. On the UAVDark135 benchmark, its precision, normalized precision and AUC surpass state-of-the-art (SOTA) methods by 5.9%, 5.4% and 4.2% respectively, while maintaining a real-time processing speed of 81 FPS. Further tests on a real-world UAV platform validate the system's reliability, demonstrating that MATrack can provide stable and effective nighttime UAV tracking support for critical robotics applications such as nighttime search and rescue and border patrol.

MATrack: Efficient Multiscale Adaptive Tracker for Real-Time Nighttime UAV Operations

TL;DR

The paper tackles the critical problem of reliable real-time nighttime UAV tracking, where low light, clutter, and dynamic viewpoints degrade performance. It introduces MATrack, a multiscale adaptive tracker built from three synergistic modules: Multiscale Hierarchy Blender for robust cross-scale fusion, Adaptive Key Token Gate for dynamic emphasis of object signals, and Nighttime Template Calibrator for stable long-term template updates, all operating on an Overlapped ViT backbone. MATrack achieves state-of-the-art results across multiple nighttime benchmarks, including UAVDark135, NAT2024-1, NAT2021, and DarkTrack2021, while maintaining real-time speed at 81 FPS. Real-world UAV tests confirm its robustness and practicality for missions such as nighttime search and rescue and border patrol, highlighting its potential for deployment in resource-constrained robotic systems.

Abstract

Nighttime UAV tracking faces significant challenges in real-world robotics operations. Low-light conditions not only limit visual perception capabilities, but cluttered backgrounds and frequent viewpoint changes also cause existing trackers to drift or fail during deployment. To address these difficulties, researchers have proposed solutions based on low-light enhancement and domain adaptation. However, these methods still have notable shortcomings in actual UAV systems: low-light enhancement often introduces visual artifacts, domain adaptation methods are computationally expensive and existing lightweight designs struggle to fully leverage dynamic object information. Based on an in-depth analysis of these key issues, we propose MATrack-a multiscale adaptive system designed specifically for nighttime UAV tracking. MATrack tackles the main technical challenges of nighttime tracking through the collaborative work of three core modules: Multiscale Hierarchy Blende (MHB) enhances feature consistency between static and dynamic templates. Adaptive Key Token Gate accurately identifies object information within complex backgrounds. Nighttime Template Calibrator (NTC) ensures stable tracking performance over long sequences. Extensive experiments show that MATrack achieves a significant performance improvement. On the UAVDark135 benchmark, its precision, normalized precision and AUC surpass state-of-the-art (SOTA) methods by 5.9%, 5.4% and 4.2% respectively, while maintaining a real-time processing speed of 81 FPS. Further tests on a real-world UAV platform validate the system's reliability, demonstrating that MATrack can provide stable and effective nighttime UAV tracking support for critical robotics applications such as nighttime search and rescue and border patrol.
Paper Structure (18 sections, 13 equations, 4 figures, 5 tables)

This paper contains 18 sections, 13 equations, 4 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Single Object Tracking
  4. Nighttime UAV Tracking
  5. Methodology
  6. Overview
  7. Multiscale Hierarchy Blender
  8. Adaptive Key Token Gate
  9. Nighttime Template Calibrator
  10. Prediction Head and Training Loss
  11. Experiment
  12. Implementation Details
  13. Comparison Results
  14. Efficiency analysis
  15. Ablation Study
  16. ...and 3 more sections

Figures (4)

  • Figure 1: This scatter plot illustrates the balance between tracking speed (FPS) and precision (%) for various methods on NAT2024-1 fu2024prompt benchmark. The red dashed line marks the 30 FPS threshold for real-time performance. Our proposed MATrack (highlighted in red) achieves the highest precision (87.7%) while maintaining a speed (81 FPS) well above the real-time requirement, demonstrating a superior efficiency-performance trade-off compared to other state-of-the-art (SOTA) methods.
  • Figure 2: Overview architecture of MATrack. The nighttime dynamic features of the static, dynamic templates and research region are fused by Multiscale Hierarchy Blender (MHB) module. As the O-ViT backbone performs feature extraction and interaction, the Adaptive Key Token Gate (AKTG) suppresses background noise tokens while emphasizing object-related information. We also designed the Nighttime Template Calibrator (NTC) to adaptively update the dynamic template and ensure reliable long-term tracking under challenging conditions.
  • Figure 3: Qualitative comparison results of our tracker with other two latest trackers (i.e., DCPT zhu2024dcpt and DRATer li2025darter) in representative nighttime scenarios. Our method maintains its robustness even in complex environments. Better viewed in color with zoom-in.
  • Figure 4: The reliability of our system is validated through real-world UAV platform tests in nighttime tracking scenarios. The frame-wise performance, represented by Center Location Error (CLE) plots, demonstrates that our tracker’s errors are consistently below the green dashed line (CLE = 20 pixels), which is the threshold for acceptable performance.