Spaceborne differential photogrammetry for control-free measurement of large-gradient deformation with structural immunity and a predictable accuracy envelope

Yueqiang Zhang; Chang Ma; Shuixin Pan; Haibo Liu

cs.CVAug 26, 2026

Spaceborne differential photogrammetry for control-free measurement of large-gradient deformation with structural immunity and a predictable accuracy envelope

Authors: Yueqiang Zhang, Chang Ma, Shuixin Pan, Haibo Liu

Organizations: State Key Laboratory of Radio Frequency Heterogeneous Integration; Key Laboratory of Optoelectronic Devices and Systems of Ministry of Education and Guangdong Province; Shenzhen Key Laboratory of Intelligent Optical Measurement and Detection; College of Physics and Optoelectronic Engineering, Shenzhen University, Shenzhen, 518060, China · School of Artificial Intelligence and Robotics, Hunan University, Changsha, 410082, China

Abstract

Optical satellite image correlation measures wide-area deformation in regimes where coherent interferometric synthetic aperture radar fails because displacement gradients are too large. However, standard pairwise workflows lack a pre-acquisition error budget and rely on extensive stable terrain. We formulate repeat-pass optical correlation as a differential estimation problem without surveyed ground control. Nominal georeferencing defines the coordinate frame, stable-area constraints and displacement priors resolve the datum, and surface displacement is estimated jointly with inter-epoch revisit-bias coefficients. The model yields a predictive accuracy envelope and calibrated per-point posterior uncertainty, bounds along-track uncertainty through a displacement prior, and represents pushbroom jitter using per-line revisit offsets. Simulations and Sentinel-2 and WorldView-2 experiments on the 2019 Ridgecrest earthquake, the 2023 Kahramanmaraş earthquake, and the Baltoro glacier validate the predicted noise floor, control-free accuracy margin, and leakage caused by view-angle and digital elevation model errors. The measured noise floor reaches approximately $0.05$ pixel at $10$ ,m ground sampling distance. With only five stable tiles, conventional destriping changes the estimated Baltoro trunk velocity from $106$ to $1251$ myr $^{-1}$ , whereas the prior-constrained estimate remains $87$ myr $^{-1}$ . Closure analysis attributes approximately $88\%$ of pair-error variance to individual scenes, consistent with $25{,}354$ ITS_LIVE glacier-velocity triplets. Three matching methods lead to the same conclusions. The framework therefore turns pairwise correlation into a robust measurement with a predictive error budget, reduced dependence on stable terrain, and conclusions independent of the matching method.

Figures & tables

Line jitter	ground/line	low-order $\xi$	per-line $\xi(t)$
$0.05^{\prime\prime}$	12 cm	17.3 cm	9.6 cm
$0.20^{\prime\prime}$	48 cm	64.0 cm	9.6 cm
$1.00^{\prime\prime}$	242 cm	318.4 cm	9.6 cm

Table 1: E2: line-jitter absorption by revisit-offset order (along-track floor).

Method	RMSE (cm)
same-name differencing	584.6
$+$ low-order (RPC-style) correction	64.6
$+$ per-line destriping (standard)	5.2
informed differential (per-line $\xi$ , $\sigma_{d}$ prior)	9.6

Table 2: Comparison of the engineering baseline and the informed differential estimator using along-track RMSE for the sparse-slip scene.

Figure 1: Ridgecrest 2019 coseismic east–west displacement obtained from the control-free Sentinel-2 pair. The left panel shows the destriping baseline, and the right panel shows the informed differential solution using the automatic prior rule described in Section 6 . Black dots denote the USGS field observations of surface rupture. The lobed coseismic pattern and the discontinuity along the mapped rupture are recovered in a single pass; the salt-flat decorrelation zone (upper right) is flagged for masking.

Figure 2: Ridgecrest

M_{\mathrm{w}}\,7.1

: image-derived cross-fault step versus field-measured near-fault offset for all 45 deduplicated profiles with field offset

>0.5

m. The relation is unbiased (ratio median 1.00); scatter grows toward the noise floor.

Figure 3: Along-strike slip distribution: image-derived steps (red) track the field measurements (black) over

45

km of rupture; the image aperture smooths the near-fault concentration at the slip maxima, the structure reported by field–image comparisons of this event.

Figure 4: Uncertainty calibration from two-band redundancy (Ridgecrest): empirical single-band

\sigma

versus correlation quality

q

(8 quantile bins) with a power-law summary; the shape is a plateau with a decorrelation knee.

Figure 5: Kahramanmaraş 2023 cross-view WorldView-2 pair (Narlı segment, 0.52 m): E-W (left) and N-S (right) displacement after detrending. The strong anisotropy—quiet across-look field versus noisy along-look field—is the look-direction resolution-degradation penalty of cross-view pairing, uncorrectable by any DEM (Section 5.2 ).

Direction	detrended	informed
across view-angle diff.	$0.29$ m ( $0.55$ px)	$0.14$ m ( $0.27$ px)
along view-angle diff.	$1.92$ m ( $3.69$ px)	$0.90$ m ( $1.73$ px)

Table 3: Kahramanmaraş cross-view pair: anisotropic noise floor (stable-area robust MAD

\sigma

Figure 6: Baltoro glacier, Sentinel-2A 10 m, control-free. Left: speed field from the 23 September–3 October pair (m yr -1 ; gray = bedrock stable area; triangles = ITS_LIVE centerline points). Right: centerline speed, this work (10-day and 20-day pairs) versus ITS_LIVE autumn and annual medians; the 20-day pair includes the fresh-snow scene.

Figure 7: Forward test. Left: autumn-2025 trunk speed measured from the 13–23 September 2025 pair versus the autumn-2024 field taken as a persistence forecast (dashed 1:1). Right: centerline speed for the ITS_LIVE 2022–2024 autumn median, 2024 and 2025 from this work.

		Noise floor		Signal
Scene	Kernel	heuristic	masked	heuristic	masked
Ridgecrest	PC	$0.50$ m	$0.49$ m	$3.83$	$3.86$
Ridgecrest	NCC	$0.55$ m	$0.53$ m	$4.14$	$4.09$
Ridgecrest	autoRIFT (pf. off)	$0.39$ m	$0.38$ m	$4.73$	$4.74$
Ridgecrest	autoRIFT (default)	$0.28$ m	$0.24$ m	$3.56$	$3.58$
Baltoro	PC	$0.81$ m	$0.73$ m	$58$	$106$

Table 4: Kernel

\times

post-processing on identical inputs (10 m,

64

-px chips,

32

-px step). Floor: stable-area MAD

\sigma

. NCC is an autoRIFT-style reimplementation; autoRIFT is the package (conda-forge 2.1.1), shown with its default high-pass prefilter off (signal-preserving) and on (default). Floors are in m; the signal is the cross-fault step in m (Ridgecrest) or the trunk median speed in m yr -1 (Baltoro); “heuristic” is the full-field detrend-and-destripe, “masked” the stable-area differential correction.

Figure 8: Thinning the stable area. Left: Ridgecrest, held-out stable floor as the fitting set shrinks from $50\%$ to $2\%$ of $5547$ tiles. Right: Baltoro, trunk median speed as the $54$ bedrock tiles are reduced to $5$ ; horizontal lines mark the zero-stable-tile informed solution and full-field destriping.

		destriping		informed
Injected field	stable	masked	full-field	plain	robust
local patch, $2$ m, $\sigma\,5$ km	100%	0.65	0.86	0.49	0.49
	20%	0.67	0.86	0.66	0.66
	5%	0.69	0.86	0.71	0.70
smooth field, $\pm 2$ m	100% (37 tiles)	0.79	1.09	1.19	1.20
	20 tiles	2.47	1.10	1.19	1.20

Table 5: Injected deformation on the real Ridgecrest background: recovery RMSE (m) against known truth. Stable fraction = share of the truly stable tiles made available for fitting; 8 random draws where

<100\%

Figure 9: Natural cut-outs of the Baltoro AB field, nothing thinned. (a) Windows with bedrock in one corner: trunk median speed (log) for plane-plus-strip destriping, constant-plus-strip destriping and the informed estimator; black tick = full-scene 54-tile masked solution. (b) Bedrock-free windows: per-tile RMS of the field shape (common median removed) against the full-scene solution.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Figure S1: The same calibration on four scene pairs. The relation is monotone everywhere; its coefficients are scene-specific (exponents $0.24$ – $0.65$ ), so $\sigma(q)$ is calibrated per pair from the two-band redundancy rather than transferred as a constant.

Figure S3: CBERS-04A 2 m (left) and Sentinel-2 10 m (right) dune migration over the same window; gray = vegetated stable area.

field / stable	$3$	$10$	$20$	$50$	$100$ m
local patch, 100%	0.59	0.48	0.46	0.46	0.46
local patch, 5%	0.76	0.69	0.67	0.67	0.67
smooth field, 100%	1.20	1.19	1.18	1.12	0.99
smooth field, 5%	1.20	1.19	1.17	1.13	1.03
strip band, 100%	1.40	1.19	0.97	0.82	0.79
strip band, 5%	1.45	1.32	1.09	0.82	0.76

Appendix

Table S6: Injected-truth RMSE on the real Ridgecrest background for different displacement-prior scales. All values are in meters.

Correction	E	N	mean (px)
raw (constant removed)	$1.390$	$3.429$	$1.205$
full-field destripe, $12^{\circ}$ (S2 value)	$0.251$	$0.467$	$0.179$
masked differential, $12^{\circ}$	$0.241$	$0.478$	$0.180$
full-field destripe, $7.5^{\circ}$ (estimated)	$0.162$	$0.311$	$0.118$

Appendix

Table S11: CBERS-04A 204/117, 28 October to 28 November 2025,

2

m, held-out stable nodes. Floors in m.

Figure S11: CBERS-04A 204/117. (a) Fraction of the detrended residual explained by strip medians against assumed strip azimuth; the maximum is at $7.5^{\circ}$ , not at the Sentinel-2 value of $12^{\circ}$ . (b) Held-out floors: removing a constant only, destriping at the Sentinel-2 azimuth, and destriping at the estimated azimuth.

Remote Sensing

Remote sensing uses satellite, aerial, and ground-based observations to measure and map Earth and other planetary surfaces.

See all 522 papers

Aug 14, 2026

Beyond Control Points: Arcsecond Relative-Motion Estimation of Vision Measurement Platforms With Incomplete or Absent Control Fields

Long-range vision-based deformation monitoring is highly sensitive to motion of the camera platform. Absolute-pose differencing typically relies on dedicated control data and propagates two independent pose errors into the relative-motion estimate. We develop a control-adaptive differential framework that estimates inter-frame platform motion directly from image displacements and known 3D points. With no dedicated control point, the framework recovers platform rotation from measurement-point observations. One surveyed control point enables prior-constrained translation recovery, while two nonparallel control rays recover full 3D translation. The framework requires neither nonlinear optimization nor an initial pose estimate. Excluding control data from the rotation stage makes the rotation estimate exactly immune to contamination confined to the control field. The inherited differential formulation also cancels translational extrinsic errors exactly. We derive the rotation observability condition, a leakage bound for unmodeled translation and nonrigid point motion, and the single-point axial-prior bias law. Under 0.5-pixel image noise, attitude changes of up to 30~~arcmin, and 3D point perturbations of up to 2~~mm, the multi-camera estimator achieves a rotation RMSE of 2.97~~arcsec and an average runtime of 0.46~~ms. With one surveyed control point, its prior-constrained translation RMSE is 1.19~~mm. In a bridge experiment without a stable control field, the median coordinate-wise displacement RMSE relative to total-station measurements is 0.85~~mm. The estimator also maintains zero divergence under the tested 3D coordinate perturbations on public RGB-D and stereo sequences. These results establish state-of-the-art accuracy, calibration robustness, and computational efficiency among the evaluated methods.

Meng Lian, Jian Wang, Shuixin Pan +3

Ministry of Natural Resources (MNR) Key Laboratory for Geo-Environmental Monitoring of Great Bay Area & Guangdong Key Laboratory of Urban Informatics, Shenzhen University, Shenzhen, China · Shenzhen Expressway Co., Ltd., Shenzhen, China · Key Laboratory of Optoelectronic Devices and Systems of Ministry of Education and Guangdong Province, the Shenzhen Key Laboratory of Intelligent Optical Measurement and Detection, and the College of Physics and Optoelectronic Engineering, Shenzhen University, Shenzhen 518060, China +2

Relative Pose Estimation Camera Pose Estimation Structural Health Monitoring

Ask PDF

Aug 23, 2026

Self-Calibrating Dense Displacement Fields for Reliable Co-Registration of Large Optical Satellite Imagery

Co-registration underlies nearly every multi-temporal and multi-sensor use of optical satellite imagery, and operational products still carry documented offsets well above the fraction-of-a-pixel scale at which change detection, time series, and data fusion degrade. Real image pairs differ along several axes at once (sensor response, scene content, viewing geometry, resolution, mosaic seams), and the last of these is not a single global motion. Existing tools embed a motion model and constants tuned to their development data; a pair that fits is registered precisely, while one that does not either fails to match or returns a result wrong by tens of pixels with no failure reported. Learned matchers add a GPU requirement and carry no accuracy guarantee outside their training distribution. We present SCDF (self-calibrating displacement fields), a training-free, GPU-free estimator whose motion model is the dense per-pixel displacement field itself, so no scene motion falls outside the model. A single predict--measure--filter loop runs over a resolution pyramid: the accumulated field predicts where each patch of the moving image falls in the reference, RootSIFT matching and a correlation pass measure the displacement there to sub-pixel precision, and filters whose thresholds are all calibrated on the image pair itself decide what survives. One configuration, with no per-dataset tuning, processes full

8192^2

scenes on a single CPU core. On 584 constructed-ground-truth pairs built from real Sentinel-2, Landsat-8/9, and NAIP imagery, against seven classical baselines and two zero-shot pretrained matchers, SCDF registers every pair with zero failures, reduces the best baseline's real-pair median end-point error from 6.83 to 4.17m, and cuts its 90th percentile from 17.8 to 7.77m.

Shoukun Sun, Zhe Wang, Sanaz Salati +3

Department of Computer Science, University of Idaho, Moscow, ID 83844 USA · National Center for Ecological Analysis and Synthesis (NCEAS), University of California, Santa Barbara, CA 93101 USA · Research Computing and Data Services (RCDS), University of Idaho, Moscow, ID 83844 USA +1

Deformable Image Registration Image Matching Remote Sensing

Ask PDF

Apr 11, 2026

Are Pretrained Image Matchers Good Enough for SAR-Optical Satellite Registration?

Cross-modal optical--SAR (Synthetic Aperture Radar) registration is a bottleneck in remote-sensing disaster response. Modern image matchers are developed and benchmarked almost exclusively on natural-image domains. We evaluate twenty-four pretrained matcher configurations in a zero-shot setting, with no fine-tuning or domain adaptation on satellite or SAR data. The evaluation spans SpaceNet9 and two additional cross-modal benchmarks under a deterministic protocol that uses tiled large-image inference, robust geometric filtering, and tie-point-grounded metrics. Our results show uneven transfer: matchers with explicit cross-modal training do not uniformly outperform those without it. XoFTR (trained for visible--thermal matching) and RoMa achieve the lowest reported mean tie-point error at

3.0

px on the labeled SpaceNet9 training scenes. RoMa achieves this result without any cross-modal training. MatchAnything-ELoFTR (

3.4

px), trained on synthetic cross-modal pairs, is close behind. XoFTR also runs roughly an order of magnitude faster than the RoMa family (

0.4

s vs.

5.0

s per pair). RoMa's performance is consistent with the hypothesis that frozen DINOv2 features confer robustness to large appearance shifts. Deployment protocol choices (geometry model, tile size, inlier gating) change mean error by up to

33\times

for a single matcher. This shift can exceed the effect of swapping matchers within our evaluated protocol ablations. Affine geometry alone reduces mean error from

12.3

9.7

px.

Isaac Corley, Alex Stoken, Gabriele Berton

Taylor Geospatial · Independent Researcher

Cross-Modal Image Registration Cross-Modal Feature Matching Cross-View Image Matching Remote Sensing Remote Sensing Benchmarks

Ask PDF

Abstract

Remote Sensing

Explore similar work

Beyond Control Points: Arcsecond Relative-Motion Estimation of Vision Measurement Platforms With Incomplete or Absent Control Fields

Self-Calibrating Dense Displacement Fields for Reliable Co-Registration of Large Optical Satellite Imagery

Are Pretrained Image Matchers Good Enough for SAR-Optical Satellite Registration?