A SIFT-Based DEM Extraction Approach Using GEOEYE-1 Satellite Stereo Pairs

Area of Science:

• Photogrammetry and Remote Sensing
• Geomatics Engineering

Background:

• Very High Resolution (VHR) satellite stereo-pair imagery offers significant potential for detailed topographic mapping.
• Processing large VHR datasets for Digital Elevation Model (DEM) extraction presents computational challenges.

Purpose of the Study:

• To present a novel module for VHR satellite stereo-pair imagery processing and DEM extraction.
• To address the challenges of handling large VHR image file sizes and ensure accurate DEM generation.

Main Methods:

• Utilized parallel processing of cascading image tiles to manage large VHR imagery.
• Employed Scale-Invariant Feature Transform (SIFT) for feature matching and RANdom SAmple Consensus (RANSAC) for filtering feature pairs.
• Extracted point cloud ground coordinates from homologous pairs and introduced average point spacing irregularity to assess DEM resolution.

Main Results:

• Successfully processed a 0.5 m × 0.5 m Geoeye-1 stereo pair, generating a 1.5-m resolution DEM.
• The resulting DEM exhibited superior detail compared to reference DEMs.
• Achieved a Root Mean Square Error (RMSE) of approximately 1 meter against ground truth measurements.

Conclusions:

• The developed module is effective for VHR satellite stereo-pair processing and DEM extraction.
• The parallel processing approach efficiently handles large datasets.
• The generated DEMs possess high accuracy and detail, suitable for various geomatic applications.