TL;DR
This paper demonstrates the feasibility of using China's Gaofen-1 Wide-Field-View (WFV) satellite imagery for wide-area stereo mapping, achieving a significant improvement in Digital Surface Model (DSM) accuracy from over 100m to approximately 30m after applying a novel calibration method. This advancement enables rapid topographic map updates and supports 1:250,000 scale mapping, overcoming limitations of previous stereo mapping techniques.
What Problem Does This Paper Address?
Traditional optical satellite stereo mapping methods, such as multiline array and agile stereo modes, face a trade-off between spatial coverage and temporal resolution. Multiline array systems offer wide coverage but have long revisit periods, limiting their use for rapid map updates. Agile stereo systems provide high temporal resolution but cover very small areas. This gap hinders applications requiring both wide-area coverage and frequent updates, like global change detection or rapid topographic map revisions. The paper addresses this by exploring the "wide swath stereo mode" using Gaofen-1 WFV imagery, which offers a broad 800 km swath and a short 4-day revisit period, aiming to achieve reliable stereo mapping accuracy despite inherent nonlinear system distortions in the WFV images.
Methods
The core of the paper's methodology lies in addressing the nonlinear system distortions present in the Gaofen-1 WFV images, which are crucial for accurate stereo mapping.
- Wide Swath Stereo Mode: This mode utilizes a single camera to acquire stereo pairs from different orbits, leveraging the wide swath of the WFV sensor. This contrasts with traditional methods that rely on multiple cameras or satellite agility.
- Nonlinear System Distortion Calibration: A key contribution is the application of a multicalibration image method, building upon previous work [25]. This method accounts for both interior (lens distortion) and exterior (attitude) errors. It involves:
- Collecting calibration images at different times to cover various image rows with Ground Control Points (GCPs).
- Using a modified polynomial model to correct for nonlinear distortions.
- Employing an affine model for exterior orientation, which is robust for images with significant overlap.
- 3D Stereo Model Generation: After calibration and orientation, a Digital Surface Model (DSM) is generated using the Semi-Global Matching (SGM) method [29] to find corresponding points between stereo image pairs. Forward intersection is then used to reconstruct the 3D object space coordinates.
Datasets
The study utilizes data from the Gaofen-1 (GF-1) satellite, specifically its Wide-Field-View (WFV) cameras.
- Calibration Images: Multiple GF-1 WFV-1 and WFV-4 images were collected over Shanxi and Henan provinces in China. These images were used to derive the calibration parameters.
- Stereo Mapping Images: Two specific GF-1 WFV images (WFV-1 scene 068316 and WFV-4 scene 112159) covering the Shanxi province were used for stereo mapping experiments.
- High-Accuracy Geometric Calibration Field (GCF): A GCF with horizontal and elevation accuracies of 1.6 m and 1.5 m, respectively, was used for validating the elevation accuracy of the generated DSMs.
Key Insights and Ideas
- Wide Swath Stereo is Feasible: The paper convincingly demonstrates that the wide swath stereo mode, despite its inherent challenges, is a viable approach for generating accurate topographic data.
- Calibration is Paramount: The significant improvement in DSM accuracy (from >100m to ~30m) highlights that effective calibration of nonlinear system distortions is not just beneficial but essential for achieving reliable results with GF-1 WFV imagery.
- Bridging the Gap: This work effectively bridges the gap between high-resolution but narrow-swath imagery and low-resolution but wide-swath imagery, offering a path towards more frequent and comprehensive topographic mapping.
- WFV-1 and WFV-4 Synergy: The study leverages the complementary coverage of WFV-1 and WFV-4 cameras to create stereo pairs, showcasing the potential of using different sensors within the same satellite system for stereo applications.
Key Results
- Orientation Accuracy Improvement: After calibration, the maximum orientation errors were reduced to less than two pixels for both WFV-1 and WFV-4 cameras, with some scenes achieving accuracies of approximately 0.5 pixels. This is a substantial improvement from pre-calibration errors of up to 11 pixels.
- DSM Elevation Accuracy: The elevation accuracy of the generated DSM was significantly improved from an average RMS error of 103.850 m before calibration to 30.116 m after calibration.
- Meeting Mapping Standards: The achieved elevation accuracy of approximately 30 m meets the requirements for 1:250,000 scale mapping and rapid topographic map updates.
Limitations and Future Work
- Resolution and Radiation Quality: The authors acknowledge that the current 16 m nadir resolution and relatively poor radiation quality of the GF-1 WFV images limit the ultimate elevation accuracy.
- Future Work Suggestion: They suggest that utilizing higher-resolution wide-swath images with improved radiation qualities, combined with proper calibration, would yield even better results for the wide swath stereo mapping mode.
Summary updated