CLEAR-IR: Clarity-Enhanced Active Reconstruction of Infrared Imagery

TL;DR

The paper tackles robust robotic perception in dark environments by leveraging active infrared (IR) streams and removing structured-light emitter patterns that impair high-level vision tasks. It introduces CLEAR-IR, a U-Net-based reconstruction framework that optimizes a composite loss $\mathcal{L}_{total} = \alpha\mathcal{L}_{mae} + \beta\mathcal{L}_{ssim} + \gamma\mathcal{L}_{freq} + \delta\mathcal{L}_{sobel} + \epsilon\mathcal{L}_{perceptual} + \zeta\mathcal{L}_{tv}$ to suppress emitter noise while preserving perceptual fidelity, trained on a 50,000-pair IR–greyscale RGB dataset augmented from 10,000 originals. The method yields a denoised IR representation that substantially improves ORB feature matching, zero-shot YOLOv8 object detection, and ArUco marker detection, outperforming several state-of-the-art low-light RGB enhancement approaches. This work demonstrates that IR-based perception, when preprocessed by CLEAR-IR, can robustly support downstream robotics tasks across illumination conditions, offering a practical preprocessing step for existing perception pipelines in autonomous systems.

Abstract

This paper presents a novel approach for enabling robust robotic perception in dark environments using infrared (IR) stream. IR stream is less susceptible to noise than RGB in low-light conditions. However, it is dominated by active emitter patterns that hinder high-level tasks such as object detection, tracking and localisation. To address this, a U-Net-based architecture is proposed that reconstructs clean IR images from emitter-populated input, improving both image quality and downstream robotic performance. This approach outperforms existing enhancement techniques and enables reliable operation of vision-driven robotic systems across illumination conditions from well-lit to extreme low-light scenes.

Figures (8)

• Figure 1: Low-light room captured using different cameras
• Figure 2: The key factors influencing the appearance of the structured light emitter pattern in active infrared imagery.
• Figure 3: The U-Net architecture used for image reconstruction and emitter pattern removal.
• Figure 4: A qualitative comparison of the raw IR input with emitter patterns, the reconstructed image, and the corresponding ground-truth image.
• Figure 5: Visual comparison of a ground truth RGB scene, a dark RGB scene, and the outputs of various image enhancement methods, including the proposed Denoised IR method.