TL;DR
This paper presents a novel method to compute digital elevation models (DEMs) from airborne Fourier Transform Spectrometer (FTS) images. This is crucial for correcting interferograms, which are often flawed in non-flat terrain, enabling more accurate spectral analysis of such areas.
What Problem Does This Paper Address?
Airborne imaging Fourier Transform Spectrometers (FTS) capture spectral information by taking a sequence of images over time. While this allows for spectral reconstruction, it introduces a challenge: when the instrument is pointed at the nadir (straight down) and the scene is not flat (e.g., has buildings or slopes), the changing view angle during acquisition leads to miscalculations in matching scene elements across images. This results in erroneous interferograms, hindering accurate spectral analysis in these non-flat areas. The paper addresses the need for accurate ground elevation estimation to identify and correct these flawed interferograms.
Methods
The authors propose a two-step method:
- Fringe Removal: Interference fringes are superimposed on the FTS images and are essential for spectrum reconstruction. However, they can interfere with relief estimation. The paper proposes removing these fringes by treating them as band-limited in the Fourier space. They employ a Total Variation (TV) minimization approach, inspired by image decomposition techniques, to separate the scene image from the fringe image. The objective is to minimize the TV of the scene image while ensuring that its Fourier transform matches the observed image's Fourier transform in the relevant frequency band.
- Depth Computation: For depth estimation, the paper leverages the Epipolar-Planes Images (EPI) method. This technique requires images to be in epipolar geometry, meaning corresponding scene elements align vertically. The authors construct EPIs by stacking columns from consecutive images. In an EPI, a scene element's path appears as a line, and the slope of this line is proportional to the object's distance from the camera. A robust line detection algorithm based on pixel gradient orientation is then used to determine these slopes and, consequently, the depth of scene elements.
Datasets
Not discussed in the paper. The paper uses data from the "Sieleters" airborne infrared imaging Fourier transform spectrometer developed by ONERA.
Key Insights and Ideas
- Fringes as a Hindrance: While crucial for spectral reconstruction, the interference fringes themselves can corrupt relief estimation if not handled properly.
- TV Minimization for Decomposition: Total Variation minimization proves effective for decomposing the observed image into a scene component and a fringe component, a common technique in image processing adapted here for a specific FTS challenge.
- EPIs for Depth: The Epipolar-Planes Image (EPI) approach, a standard in multi-view geometry, is effectively applied to FTS sequences for depth estimation.
- Slope-Depth Proportionality: The core idea of EPIs – that the slope of a scene element's trajectory in an EPI is directly related to its depth – is fundamental to their depth computation.
- Altitude Dominance: The paper notes that due to the aircraft's high altitude, elevation differences in the scene have a less pronounced effect on the EPI slopes compared to the overall scene geometry, simplifying the depth estimation process.
Key Results
The paper successfully demonstrates the computation of a sparse elevation map from FTS images. This elevation map is shown to be useful for identifying flawed interferograms in non-flat areas. The method allows for more accurate interferogram computation in such regions, which is a significant step towards reliable spectral analysis of complex terrains. Figure 4 illustrates a sparse elevation map overlaid on the original image, showing height coded in color.
Limitations and Future Work
Limitations: * The method produces a sparse elevation map, meaning it doesn't provide depth for every pixel. * The paper acknowledges that some artifacts may still be visible in areas with high radiance gradients after fringe removal.
Future Work: * The authors plan to explore precise interpolation techniques to densify the sparse elevation map, potentially creating a full digital elevation model for all scene points. * Further work will focus on precise interpolation of interferograms on irregular sampling grids, building upon the corrected interferograms.
Summary updated