Lightweight Multispectral Detection and DEM-Constrained Ray Consistency Localization for UAV-Assisted Search and...

Area of Science:

• Robotics and Automation
• Computer Vision
• Geospatial Analysis

Background:

• Unmanned aerial vehicle (UAV)-assisted search and rescue (SAR) operations require precise target detection and localization, which are hindered by complex outdoor environments.
• Challenges include background clutter, occlusion, low illumination, and infrared thermal diffusion affecting small targets in Red-Green-Blue-Infrared (RGB-IR) imagery.
• Unstable viewpoints and terrain variations introduce uncertainty in geographic localization.

Purpose of the Study:

• To develop an integrated UAV-SAR framework for enhanced small target detection and accurate geographic localization.
• To address limitations in current UAV-SAR systems concerning target visibility and positional accuracy in diverse environmental conditions.

Main Methods:

• Introduced the Asymmetric Fusion and Context-aware Detection (AFC-Det) network for lightweight multispectral detection using RGB-IR data.
• Developed the Global Context-Regularized Huber Ray Consistency Optimization (GCR-HRCO) for Digital Elevation Model (DEM)-constrained geographic localization.
• Employed asymmetric dual-stream encoding, cross-modal mutual prompting, high-resolution anchored aggregation, global ray aggregation, multi-ray geometric consistency, Huber robust optimization, and DEM-based terrain constraints.

Main Results:

• AFC-Det achieved 45.4% average precision (AP) and 44.7% AP for small objects (APs) on the VTSaR dataset with high efficiency (1.7M parameters, 107.2 FPS).
• AFC-Det demonstrated good generalization to the M3FD dataset, achieving 54.6% AP.
• GCR-HRCO significantly reduced mean horizontal error (from 6.85 m to 3.31 m) and RMSE (from 8.16 m to 4.33 m) on the SAR-DAG_raycast dataset.

Conclusions:

• The proposed AFC-Det network effectively enhances small-target representation from RGB-IR imagery for UAV-SAR.
• The GCR-HRCO method significantly improves geolocation accuracy by integrating DEM data and robust optimization techniques.
• The integrated framework demonstrates a substantial advancement in UAV-assisted search and rescue capabilities.