Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Simple Detector with Frame Dynamics is a Strong Tracker

Chenxu Peng, Chenxu Wang, Minrui Zou, Danyang Li, Zhengpeng Yang, Yimian Dai, Ming-Ming Cheng, Xiang Li

TL;DR

The paper tackles infrared tiny-object tracking for Anti-UAV by turning a strong object detector into a robust tracker through frame dynamics and trajectory-based post-processing. By augmenting inputs with frame difference and optical flow, the model learns both appearance and motion priors, while TC-Filtering enforces spatiotemporal consistency to suppress false positives. The approach selects strong infrared detectors, fuses temporal cues, and applies motion-based constraints, achieving state-of-the-art results in the 4th Anti-UAV Challenge (1st in Track 1, 2nd in Track 2). This yields a practical framework for reliable tiny-target tracking in challenging infrared UAV scenarios.

Abstract

Infrared object tracking plays a crucial role in Anti-Unmanned Aerial Vehicle (Anti-UAV) applications. Existing trackers often depend on cropped template regions and have limited motion modeling capabilities, which pose challenges when dealing with tiny targets. To address this, we propose a simple yet effective infrared tiny-object tracker that enhances tracking performance by integrating global detection and motion-aware learning with temporal priors. Our method is based on object detection and achieves significant improvements through two key innovations. First, we introduce frame dynamics, leveraging frame difference and optical flow to encode both prior target features and motion characteristics at the input level, enabling the model to better distinguish the target from background clutter. Second, we propose a trajectory constraint filtering strategy in the post-processing stage, utilizing spatio-temporal priors to suppress false positives and enhance tracking robustness. Extensive experiments show that our method consistently outperforms existing approaches across multiple metrics in challenging infrared UAV tracking scenarios. Notably, we achieve state-of-the-art performance in the 4th Anti-UAV Challenge, securing 1st place in Track 1 and 2nd place in Track 2.

A Simple Detector with Frame Dynamics is a Strong Tracker

TL;DR

The paper tackles infrared tiny-object tracking for Anti-UAV by turning a strong object detector into a robust tracker through frame dynamics and trajectory-based post-processing. By augmenting inputs with frame difference and optical flow, the model learns both appearance and motion priors, while TC-Filtering enforces spatiotemporal consistency to suppress false positives. The approach selects strong infrared detectors, fuses temporal cues, and applies motion-based constraints, achieving state-of-the-art results in the 4th Anti-UAV Challenge (1st in Track 1, 2nd in Track 2). This yields a practical framework for reliable tiny-target tracking in challenging infrared UAV scenarios.

Abstract

Infrared object tracking plays a crucial role in Anti-Unmanned Aerial Vehicle (Anti-UAV) applications. Existing trackers often depend on cropped template regions and have limited motion modeling capabilities, which pose challenges when dealing with tiny targets. To address this, we propose a simple yet effective infrared tiny-object tracker that enhances tracking performance by integrating global detection and motion-aware learning with temporal priors. Our method is based on object detection and achieves significant improvements through two key innovations. First, we introduce frame dynamics, leveraging frame difference and optical flow to encode both prior target features and motion characteristics at the input level, enabling the model to better distinguish the target from background clutter. Second, we propose a trajectory constraint filtering strategy in the post-processing stage, utilizing spatio-temporal priors to suppress false positives and enhance tracking robustness. Extensive experiments show that our method consistently outperforms existing approaches across multiple metrics in challenging infrared UAV tracking scenarios. Notably, we achieve state-of-the-art performance in the 4th Anti-UAV Challenge, securing 1st place in Track 1 and 2nd place in Track 2.
Paper Structure (14 sections, 5 equations, 5 figures, 5 tables, 1 algorithm)

This paper contains 14 sections, 5 equations, 5 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Object Detection
  4. Object Tracking
  5. Evolving a Detector into a Tracker
  6. Selecting a base detector
  7. Empowering detector with frame dynamics
  8. Refining tracking with TC-Filtering
  9. Experiments
  10. Implementation Details
  11. Quantitative Evaluation
  12. Qualitative Evaluation
  13. Ablation Study
  14. Conclusion

Figures (5)

  • Figure 1: Visualization of heatmaps from different detection models. The first row shows the original images, while the second to fourth rows correspond to the detection results of different types of model inputs: the original image, the original image concatenated with the frame difference map, and the original image concatenated with the optical flow map. The heatmaps are generated using the HiResCAM draelos2020use method.
  • Figure 2: Comparison between the conventional object trackers and the proposed object tracker. Figure (c) only utilizes the frame difference method as an example, while the optical flow method follows the same approach.
  • Figure 3: Visualization of frame difference maps and optical flow maps. The first row shows the original images. The second and third rows display the frame difference maps between the previous frame and the current frame, as well as between the frame before last and the current frame, respectively. The fourth and fifth rows present the optical flow maps in the horizontal and vertical directions between the previous frame and the current frame. Each column represents a different scene. Note that the optical flow maps are actually grayscale images, but colors have been added here for better visualization.
  • Figure 4: The visualization of our method's inference results on the test set. Each row represents a different video sequence.
  • Figure 5: (a) Impact of the sampling stride on detection performance. (b) The impact of IoU threshold, confidence threshold, and the maximum number of detection boxes per image on detection performance.