Depth-Guided RGB-D Detection for Tiny Pavement Debris in Complex Warehouse Inspection Scenes

Abstract Autonomous guided vehicles (AGVs) are increasingly used for inspection and transportation tasks in aviation warehouses. However, dense shelving, local shadows, and uneven illumination often degrade RGB image quality, reducing the reliability of vision-based perception. Tiny pavement debris, such as packaging fragments, plastic bags, cigarette butts, and small hardware items, may interfere with AGV locomotion and local perception, posing risks to warehouse inspection safety. To address this problem, this paper proposes a depth-guided RGB-depth (RGB-D) detection framework, termed PDE-RT-DETR, for robust tiny debris detection in low-illumination aviation warehouse scenes. The framework introduces a depth branch to provide auxiliary geometric cues, designs a Depth Enhancement Module (DEM) to improve depth-related feature representation through nonlinear depth discretization and re-encoding, and incorporates Continuous Depth Position Encoding (CDPE) to inject continuous depth-aware positional information into the Transformer feature space. A self-built RGB-D pavement debris dataset containing 5,635 aligned image pairs was constructed under multiple illumination conditions. Experimental results show that PDE-RT-DETR achieves 86.4% precision, 87.6% recall, 91.4% mAP0.50, and 66.8% mAP0.50:0.95. Compared with the RGB-D RT-DETR-R18 baseline, the proposed method improves mAP0.50 and mAP0.50:0.95 by 9.2 and 13.3 percentage points, respectively. Under extreme low illumination, PDE-RT-DETR maintains 82.1% mAP0.50, with a degradation rate of only 16.14%. These results demonstrate that depth-guided RGB-D perception improves the accuracy and robustness of tiny pavement debris detection in complex warehouse inspection scenes.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-17
DOI
https://doi.org/10.21203/rs.3.rs-11594712/v1
Primary Topic
Advanced Neural Network Applications
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Depth-Guided RGB-D Detection for Tiny Pavement Debris in Complex Warehouse Inspection Scenes

Runxia Guo, 陈玖圣 Chen Jiusheng, 刘贵行 Liu Guihang, Jiaxin Li et al.
2 links
Research Square
Advanced Neural Network Applications
preprint

Depth-Guided RGB-D Detection for Tiny Pavement Debris in Complex Warehouse Inspection Scenes

Runxia Guo, 陈玖圣 Chen Jiusheng, 刘贵行 Liu Guihang, Jiaxin Li, Jun Wu
preprint en

Abstract

Abstract Autonomous guided vehicles (AGVs) are increasingly used for inspection and transportation tasks in aviation warehouses. However, dense shelving, local shadows, and uneven illumination often degrade RGB image quality, reducing the reliability of vision-based perception. Tiny pavement debris, such as packaging fragments, plastic bags, cigarette butts, and small hardware items, may interfere with AGV locomotion and local perception, posing risks to warehouse inspection safety. To address this problem, this paper proposes a depth-guided RGB-depth (RGB-D) detection framework, termed PDE-RT-DETR, for robust tiny debris detection in low-illumination aviation warehouse scenes. The framework introduces a depth branch to provide auxiliary geometric cues, designs a Depth Enhancement Module (DEM) to improve depth-related feature representation through nonlinear depth discretization and re-encoding, and incorporates Continuous Depth Position Encoding (CDPE) to inject continuous depth-aware positional information into the Transformer feature space. A self-built RGB-D pavement debris dataset containing 5,635 aligned image pairs was constructed under multiple illumination conditions. Experimental results show that PDE-RT-DETR achieves 86.4% precision, 87.6% recall, 91.4% mAP0.50, and 66.8% mAP0.50:0.95. Compared with the RGB-D RT-DETR-R18 baseline, the proposed method improves mAP0.50 and mAP0.50:0.95 by 9.2 and 13.3 percentage points, respectively. Under extreme low illumination, PDE-RT-DETR maintains 82.1% mAP0.50, with a degradation rate of only 16.14%. These results demonstrate that depth-guided RGB-D perception improves the accuracy and robustness of tiny pavement debris detection in complex warehouse inspection scenes.

Research Square
Civil Aviation University of China (CN)
National Natural Science Foundation of China
Industry, innovation and infrastructure
Advanced Neural Network Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon