The Corona KH-4 reconnaissance satellite missions acquired panoramic stereo imagery with high spatial resolution of 1.8–7.5 m from 1962 to 1972. The potential of 800000+ declassified Corona images has not been leveraged due to the complexities arising from handling of panoramic imaging geometry, film distortions, and limited availability of the metadata required for georeferencing of the Corona imagery. This article presents the Corona Stereo Pipeline (CoSP): A pipeline for processing of Corona KH-4 stereo panoramic imagery. CoSP utilizes the deep learning-based feature matcher SuperGlue to automatically match feature points between Corona KH-4 images and recent satellite imagery to generate ground control points (GCPs). To model the imaging geometry and the scanning motion of the panoramic KH-4 cameras, a rigorous camera model consisting of modified collinearity equations with time-dependent exterior orientation parameters is employed. Using the entire frame of the Corona image, bundle adjustment with well-distributed GCPs results in an average standard deviation (SD) or $\sigma _{0}$ of less than two pixels. We evaluate fiducial marks on the Corona films and show that pre-processing the Corona images to compensate for film bending improves the 3-D reconstruction accuracy. The distortion pattern of image residuals of GCPs and $y$ -parallax in epipolar resampled images suggest that film distortions due to long-term storage likely cause systematic deviations of up to six pixels. Compared to the Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM), the Corona DEM computed using CoSP achieved a normalized median absolute deviation (NMAD) of elevation differences of $\approx $ 4 m over an area of approximately 4000 km2 after a tile-based fine coregistration of the DEMs. We further assess CoSP on complex scenes involving high relief and glacierized terrain and show that the resulting DEMs can be used to compute long-term glacier elevation changes over large areas.

Unmanned aerial vehicles (UAV) are increasingly used for topographic mapping. The camera calibration for UAV image blocks can be performed a priori or during the bundle block adjustment (self-calibration). For an area of interest with flat scene and corridor configuration, the focal length of camera is highly correlated with the height of the camera. Furthermore, systematic errors of camera calibration accumulate on the longer dimension and cause deformation. Therefore, special precautions must be taken when estimating camera calibration parameters. In order to better investigate the impact of camera calibration errors, a synthetic, error-free aerial image block is generated to simulate several issues of interest. Firstly, the erroneous focal length in the case of camera pre-calibration is studied. Nadir images are not able to prevent camera poses from drifting to compensate for the erroneous focal length, whereas the inclusion of oblique images brings significant improvement. Secondly, the case where the focal length varies gradually (e.g., when the camera subject to temperature changes) is investigated. The neglect of this phenomenon can substantially degrade the 3D measurement accuracy. Different flight configurations and flight orders are analyzed, the combination of oblique and nadir images shows better performance. At last, the rolling shutter effect is investigated. The influence of camera rotational motion on the final accuracy is negligible compared to that of the translational motion. The acquisition configurations investigated are not able to mitigate the degradation introduced by the rolling shutter effect. Other solutions such as correcting image measurements or including camera motion parameters in the bundle block adjustment should be exploited.

The Planet PlanetScope cubesat constellation acquires high-resolution optical images which cover the entire surface of the Earth daily, enabling an unprecedented capability to monitor the Earth surface changes. However, our analysis reveals artifacts of the geometry of PlanetScope images related to the imaging system and processing issues, limiting the usability of these data for various earth science applications, including the monitoring of glaciers, dune motion, or the measurement of ground deformation due to earthquakes and landslides. Here we analyze these artifacts and propose ways to remediate them. We use two examples to evaluate the data and assess the performance of our proposed approaches. The first is the ground deformation caused by the 2019 Ridgecrest earthquake sequence, California, and the second is the 2018-2019 surge of the Shisper glacier in the Karakorum. Using an image correlation technique, we show that PlanetScope images exhibit several geometric artifacts, such as scene-to-scene misregistration, inconsistence geolocation accuracy between spectral bands, and topographic artifacts. Altogether these artifacts make a quantitative analysis of ground displacement difficult and inaccurate. We present a method that remediates most of these geometric artifacts. In addition, we propose a framework for selecting the most appropriate images and a procedure for refining the Rational Function Model (RFM) of unrectified images to monitor surface displacements and topography changes in 3D. These tools should enhance the use of PlanetScope images for Earth Science applications.

With the development of unmanned aerial vehicles (UAVs) and global navigation satellite system (GNSS), the accurate camera positions at exposure can be known and the GNSS-assisted bundle block adjustment (BBA) approach is possible for integrated sensor orientation (ISO). This study employed ISO approach for camera pose determination with the objective of investigating the impact of a good sensor pre-calibration on a poor acquisition geometry. Within the presented works, several flights were conducted on a dike by a small UAV embedded with a metric camera and a GNSS receiver. The multi-lever-arm estimation within the BBA procedure makes it possible to merge image blocks of different configurations such as nadir and oblique images without physical constraints on camera and GNSS antenna positions. The merged image block achieves a better accuracy and the sensor self-calibrated well. The issued sensor calibration is then applied to a less preferable acquisition configuration and the accuracy is significantly improved. For a corridor acquisition scene of about 600 $\mathrm{m}$ , a centimetric accuracy is reached with one GCP. With the provided sensor pre-calibration, an accuracy of 3.9 $\mathrm{c}\mathrm{m}$ is achieved without any GCP.

Digital surface model generation using traditional multi-view stereo matching (MVS) performs poorly over non-Lambertian surfaces, with asynchronous acquisitions, or at discontinuities. Neural radiance fields (NeRF) offer a new paradigm for reconstructing surface geometries using continuous volumetric representation. NeRF is self-supervised, does not require ground truth geometry for training, and provides an elegant way to include in its representation physical parameters about the scene, thus potentially remedying the challenging scenarios where MVS fails. However, NeRF and its variants require many views to produce convincing scene’s geometries which in earth observation satellite imaging is rare. In this paper we present SparseSat-NeRF (SpS-NeRF) – an extension of Sat-NeRF adapted to sparse satellite views. SpS-NeRF employs dense depth supervision guided by crosscorrelation similarity metric provided by traditional semi-global MVS matching. We demonstrate the effectiveness of our approach on stereo and tri-stereo Pléiades 1B/WorldView-3 images, and compare against NeRF and Sat-NeRF. The code is available at https://github.com/LulinZhang/SpS-NeRF