Decoupled Motion Representation Learning for Moving Infrared Small Target Detection

2026-06-13 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, quick

Summary

A new decoupled motion representation learning framework has been developed to improve moving infrared small target detection in dynamic scenes. This framework addresses the challenge of highly coupled motions among targets, imaging platforms, and backgrounds, which often leads to an inherent trade-off between detection and false alarms in existing multi-frame methods. The core observation is that background motions are globally coherent, while small targets represent sparse local motion anomalies, and many false alarms originate from coherent background dynamics. The proposed solution introduces an explicit motion branch to model global coherent motion using pretrained optical flow priors and a self-supervised adaptation strategy. Concurrently, an implicit motion branch employs deformable feature alignment to capture target-sensitive local anomalies under coherent motion guidance. A coherent-motion-guided local anomaly reasoning module further suppresses false responses. Experiments on two infrared small target detection benchmarks show consistent outperformance over existing approaches, particularly in complex dynamic scenes, while maintaining favorable inference efficiency.

Key takeaway

For Computer Vision Engineers developing surveillance or defense systems, if you are struggling with high false alarm rates in infrared small target detection within dynamic scenes, this decoupled motion representation learning framework offers a significant advancement. You should consider integrating its explicit motion branch for global background coherence and its implicit branch for local target anomaly capture. This approach can substantially improve detection accuracy and reduce coherent-motion-induced false positives, especially in complex, dynamic environments, while maintaining efficient inference.

Key insights

Decoupling global background motion from local target anomalies significantly improves infrared small target detection.

Principles

• Background motions exhibit strong global coherence.
• Small targets correspond to sparse local motion anomalies.
• False alarms often align with globally coherent motion patterns.

Method

The framework uses an explicit motion branch with optical flow priors for global coherence and an implicit branch with deformable alignment for local anomalies, plus a module to suppress coherent-motion-induced false responses.

In practice

• Use pretrained optical flow priors for global motion modeling.
• Employ deformable feature alignment for local anomaly capture.
• Integrate a reasoning module to suppress coherent false positives.

Topics

• Infrared Small Target Detection
• Motion Representation Learning
• Optical Flow
• Deformable Feature Alignment
• Computer Vision
• Dynamic Scenes

Best for: Research Scientist, AI Scientist, Computer Vision Engineer

Related on AIssential

• Learning with Semantic Priors: Stabilizing Point-Supervised Infrared Small Target Detection via Hierarchical Knowledge Distillation — Hierarchical knowledge distillation with VFM semantics stabilizes point-supervised infrared small target detection.
• MoRe: Motion-aware Feed-forward 4D Reconstruction Transformer — MoRe efficiently reconstructs dynamic 4D scenes from monocular video by disentangling motion from static structure.
• RPCASSM: Robust PCA State Space Model For Infrared Small Target Detection — RPCASSM leverages Robust PCA to design specialized state space modules for background and target, accurately detecting infrared small targets by their spatial properties.
• Bounding-Box Trajectories Matter for Video Anomaly Detection — Bounding-box trajectories, a byproduct of detection and tracking, are a powerful, multi-class, and efficient primary signal for video anomaly detection.
• RISE: Single Static Radar-based Indoor Scene Understanding — RISE utilizes multipath radar reflections and generative AI for privacy-preserving, geometry-aware indoor scene understanding from a single static sensor.
• AMD-Powered 3D Gaussian Splatting for Autonomous Driving Scenes — Street Gaussians on AMD GPUs enables dynamic 3D scene reconstruction and editing for autonomous driving.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.