DeTracker: Motion-decoupled Vehicle Detection and Tracking in Unstabilized Satellite Videos

TL;DR

DeTracker tackles multi-object tracking in unstabilized satellite videos by decoupling platform-induced global motion from true object motion and by strengthening temporal representations for tiny targets. It introduces Global–Local Motion Decoupling (GLMD) to align features across frames and Local Refinement to preserve fine-grained motion cues, along with Temporal Dependency Feature Pyramid (TDFP) for cross-frame feature fusion. The approach yields substantial MOTA and IDF1 gains on both simulated (SDM-Car-SU) and real unstabilized satellite data, outperforming state-of-the-art methods and demonstrating practical viability for on-orbit service applications. The release of SDM-Car-SU provides a rigorous benchmark for evaluating robustness under diverse motion perturbations in unstabilized satellite video MOT.

Abstract

Satellite videos provide continuous observations of surface dynamics but pose significant challenges for multi-object tracking (MOT), especially under unstabilized conditions where platform jitter and the weak appearance of tiny objects jointly degrade tracking performance. To address this problem, we propose DeTracker, a joint detection-and-tracking framework tailored for unstabilized satellite videos. DeTracker introduces a Global--Local Motion Decoupling (GLMD) module that explicitly separates satellite platform motion from true object motion through global alignment and local refinement, leading to improved trajectory stability and motion estimation accuracy. In addition, a Temporal Dependency Feature Pyramid (TDFP) module is developed to perform cross-frame temporal feature fusion, enhancing the continuity and discriminability of tiny-object representations. We further construct a new benchmark dataset, SDM-Car-SU, which simulates multi-directional and multi-speed platform motions to enable systematic evaluation of tracking robustness under varying motion perturbations. Extensive experiments on both simulated and real unstabilized satellite videos demonstrate that DeTracker significantly outperforms existing methods, achieving 61.1% MOTA on SDM-Car-SU and 47.3% MOTA on real satellite video data.

Figures (8)

• Figure 1: Major challenges of object tracking in unstabilized satellite videos. (a) Motion decoupling ambiguity: platform-induced jitter introduces complex global motion, causing discrepancies between the true trajectories of ground objects and the apparent motion observed in unstabilized satellite videos; (b) Tiny-object representation: objects of interest exhibit extremely small scales and limited visual cues, leading to insufficient feature representation.
• Figure 2: Visualization of inter-frame motion and annotation mapping under unstabilized conditions. The left columns present the frame-wise annotation mapping results under different types of unstabilized perturbations, while the rightmost column visualizes the simulated platform motion directions.
• Figure 3: Overview of the proposed DeTracker. The framework consists of three components: (1) a feature extraction backbone for acquiring multi-scale spatial representations from input frames; (2) a Global–Local Motion Decoupling (GLMD) module for disentangling platform-induced global motion from target-level local motion; (3) a Temporal Dependency Feature Pyramid (TDFP) module for aggregating temporal information and modeling cross-frame dependencies.
• Figure 4: The structure of the GLMD module comprises a global alignment section on the left, responsible for capturing global changes between adjacent frames; and a local refinement section on the right, which processes local motion details based on the global alignment.
• Figure 5: Examples of detection results for vehicles moving in different directions. From the first line to the third line are frames 50, 120, and 189 respectively and vehicles undergo lateral and longitudinal motion. Green boxes denote true positives and correct predictions, red boxes indicate false positives and yellow boxes represent false negatives.