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

Nathan Shankar; Marton Gonczy; Hujun Yin; Pawel Ladosz

cs.ROOct 6, 2025

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

Authors: Nathan Shankar, Marton Gonczy, Hujun Yin, Pawel Ladosz

Organizations: The authors are with the the Mechanical, Aerospace and Civil Engineering and the Electrical and Electronics Departments at the University of Manchester, Manchester, United Kingdom, M13 9PL.

Abstract

This paper presents CLEAR-IR (Clarity Enhanced Active Reconstruction of Infrared Imagery), a novel framework for enabling robust robotic perception in dark environments using active infrared (IR) sensing. IR streams are less susceptible to noise than RGB in low light conditions but are dominated by structured light emitter patterns that hinder high level tasks such as object detection, tracking, and localisation. CLEAR-IR addresses this by reconstructing clean, RGB styled images from emitter populated IR inputs using a dual stream architecture inspired by DeepMAO, enabling standard RGB trained vision pipelines to operate without visible light. Evaluation across object detection, ArUco marker detection, and visual SLAM demonstrates that CLEAR-IR produces reliable perception in conditions ranging from well lit to completely dark, where passive RGB based enhancement methods fail entirely due to the physical absence of sensor signal.

Figures & tables

Panel of Figure 1 — Fig. 1: The same low light scene captured with RGB (left) and active IR (right). The RGB image contains no usable visual information, whereas the IR image clearly reveals scene structure despite the absence of ambient light.

Panel of Figure 2 — Fig. 2: Increasing laser power progresses from invisible dots to a fully populated pattern.

Panel of Figure 3 — Fig. 3: Key factors influencing emitter pattern appearance. (a) High reflectivity surfaces (plastic) show denser patterns than low reflectivity ones (foam). (b) Dots are larger and more intense at close range. (c,d) Ambient light suppresses effective emitter visibility and the pattern is much stronger in darkness.

Panel of Figure 4 — Fig. 4: CLEAR-IR Architecture

Panel of Figure 5 — Fig. 5: Visual comparison of enhancement methods on an unseen scene. RGB based methods (c–g) are applied to the dark passive RGB stream and exhibit noise and colour artefacts. Raw IR (h) retains emitter dot patterns throughout. The U-Net baseline (i) removes most patterns but produces some residual blurring in reflective regions. CLEAR-IR (j) produces a clean, structurally coherent reconstruction without introducing hallucinated features. See zoomed crops in supplementary material for edge region detail.

RGB based methods (passive dark RGB stream)
Method	Object Detection (Visible Objects)			ArUco Marker Detection
	Detected	Missed	Mislabelled	Correct	Missed
Retinex Theory [ 25 ]	1	9	0	5	1
CLAHE [ 24 ]	1	9	0	3	3
Zero-DCE [ 26 ]	0	10	0	5	1
LLFormer [ 27 ]	0	10	0	5	1

TABLE I: Performance comparison of image enhancement methods on high level robotics perception tasks

Panel of Figure 7 — Fig. 6: Object detection results. YOLOv26 fails entirely on raw IR (b) due to emitter dot patterns. On U-Net output (c), most objects are detected but three are missed. On CLEAR-IR output (d), all 11 objects are detected but some mislabelling occurs due to the greyscale nature of IR, but detections are consistent across frames.

Panel of Figure 8 — Fig. 7: ArUco marker detection. (a) RGB provides the reference; all 6 markers are detected. (b) Raw IR: 0/6 detected due to emitter dot corruption. (c) CIDNet: 6/6 detected (applied to passive dark RGB stream). (d) Original DeepMAO (retrained on IR dataset): 0/6 — over smoothing destroys marker grids. (e) U-Net baseline: 6/6. (f) CLEAR-IR: 6/6, with clean marker boundaries.

RGB based methods (dark RGB stream)
Method	Low Light (LL) - RMSE $\pm$ std			Extreme Low Light (ELL) - RMSE $\pm$ std			Inference (ms)
	FB	SQ1	SQ2	FB	SQ1	SQ2	Mean	std
Retinex [ 25 ]	0.064 $\pm$ 0.02	0.163 $\pm$ 0.05	0.168 $\pm$ 0.05	DNI	DNI	DNI	73.43	10.54
CLAHE [ 24 ]	0.069 $\pm$ 0.03	0.156 $\pm$ 0.05	0.169 $\pm$ 0.05	DNI	DNI	DNI	15.16	3.09
Zero-DCE [ 26 ]	0.065 $\pm$ 0.02	0.147 $\pm$ 0.05	0.190 $\pm$ 0.05	DNI	DNI	DNI	46.35	2.32
LLFormer [ 27 ]	0.067 $\pm$ 0.03	0.156 $\pm$ 0.05	0.183 $\pm$ 0.05	DNI	DNI	DNI	236.88	10.20

TABLE II: Quantitative evaluation of SLAM trajectory accuracy (RMSE in meters) and inference latency across Low Light (LL) and Extreme Low Light (ELL) scenarios.

Panel of Figure 10 — Fig. 8: SQ1 (Square 1 lap) trajectories. In low light (a), CLEAR-IR and U-Net closely follow the ground truth alongside the best RGB methods. In extreme low light (b), all RGB methods fail to initialise

Image Reconstruction

Image reconstruction recovers images, volumes, or scenes from incomplete, noisy, indirect, or compressed measurements using physical models and learned priors.

See all 287 papers

InfraNet: Quality-Aware RGB Guidance for Efficient Infrared Object Detection
Jul 4, 2026Zichao Feng, Haodong Zhu, Jingying Yang +8Efficient Multimodal Inference Multimodal Robustness
Beihang University, Beijing, China · Zhongguancun Academy, Beijing, China · Communication University of China, Beijing, China +1
FaithIR: Rethinking Infrared Image Super-Resolution from Perceptual Sharpness to Task Relevant Fidelity
Aug 4, 2026Axi Niu, Zhenguo Wu, Kang Zhang +3Remote Sensing Infrared Image Super-Resolution Image Super-Resolution
School of Computer Science, Northwestern Polytechnical University, Xi’an, Shaanxi, China · School of Electrical Engineering, KAIST, Daejeon, Republic of Korea · School of Aeronautics and Astronautics, Northwestern Polytechnical University, Xi’an, Shaanxi, China
SenseWeather: Multimodal Sensing for Robust Robotic Perception under Extreme Conditions
Oct 8, 2026Fucai Ke, Sukai Huang, Junyi Hao +4Robotic Perception Multimodal Robustness
Faculty of Information Technology, Monash University, Clayton, VIC 3800, Australia

Jul 4, 2026J/K move · Enter open · S save

InfraNet: Quality-Aware RGB Guidance for Efficient Infrared Object Detection

Zichao Feng, Haodong Zhu, Jingying Yang, Sheng Xu+7

Beihang University, Beijing, China · Zhongguancun Academy, Beijing, China · Communication University of China, Beijing, China+1

Efficient Multimodal Inference Multimodal Robustness Multispectral Object Detection Infrared Object Detection Multimodal Fusion

Open paper AskPDF arXiv

Robust object detection under adverse visual conditions remains a long-standing challenge for multi-modal perception systems. Existing fusion-based methods typically require both RGB and infrared (IR) inputs, and treat them equally during both training and inference, which compromises their robustness when the RGB modality becomes unreliable or unavailable. In this case, we propose InfraNet, an IR-centric quality-aware framework that regulates RGB guidance during training and supports flexible RGB--IR or IR-only deployment. InfraNet employs an asymmetric architecture where the primary IR pathway extracts multi-scale infrared features for predictions, while the auxiliary RGB pathway provides reliability-controlled supervisory signals. The core of InfraNet is QualGate, a quality-aware fusion module that learns a task-oriented control signal to suppress unreliable RGB guidance and compensate IR features during cross-modal training. Built upon InfraNet, we design two architectural variants: a lightweight IR-only architecture InfraNet-IR and an RGB--IR architecture InfraNet-RGB-IR. Our method is evaluated through extensive experiments on four benchmark datasets (LLVIP, FLIR-Aligned, M

^3

FD, and DroneVehicle), showing strong or competitive accuracy in challenging low-light and adverse weather conditions. Notably, InfraNet maintains high efficiency in IR-only inference, making it both accurate and computationally efficient.

Abstract

Image Reconstruction

Explore similar work

InfraNet: Quality-Aware RGB Guidance for Efficient Infrared Object Detection

FaithIR: Rethinking Infrared Image Super-Resolution from Perceptual Sharpness to Task Relevant Fidelity

SenseWeather: Multimodal Sensing for Robust Robotic Perception under Extreme Conditions