Adaptive block optimization for more reliable photogrammetry

MetricAware

MetricAware is a proprietary software technology developed to effectively reduce post-alignment reprojection error while preserving the geometry and connectivity of the photogrammetric block. MetricAware is available exclusively as a service: after sending the images, the client receives an optimized photogrammetric block ready for import into COLMAP or Metashape.
image proj_initial proj_final proj_delta rms_re_initial rms_re_final rms_re_delta
DJI_20250524112820_0127_D 1625 1519 -106 (-6.52%) 1.65 0.70 -0.95 (-57.58%)
DJI_20250524112825_0128_D 2754 2605 -149 (-5.41%) 1.38 0.70 -0.68 (-49.28%)
DJI_20250524112830_0129_D 2736 2552 -184 (-6.72%) 1.84 0.68 -1.16 (-63.04%)
DJI_20250524112835_0130_D 3074 2905 -169 (-5.50%) 1.54 0.70 -0.84 (-54.55%)
DJI_20250524112840_0131_D 3601 3433 -168 (-4.67%) 1.46 0.69 -0.77 (-52.74%)
DJI_20250524112845_0132_D 3927 3777 -150 (-3.82%) 1.33 0.69 -0.64 (-48.12%)
DJI_20250524112850_0133_D 4272 4124 -148 (-3.46%) 1.38 0.68 -0.70 (-50.72%)
DJI_20250524112855_0134_D 4415 4285 -130 (-2.94%) 1.26 0.70 -0.56 (-44.44%)
DJI_20250524112900_0135_D 4569 4436 -133 (-2.91%) 1.30 0.67 -0.63 (-48.46%)
DJI_20250524112905_0136_D 4460 4335 -125 (-2.80%) 1.21 0.69 -0.52 (-42.98%)

Uncompromising results

MetricAware performs an adaptive optimization guided by the state of the block. The table below reports two optimization outcomes on the same dataset, each with a different reprojection error target. The values show that the optimization significantly reduced the RMS error while keeping the number of tie points virtually unchanged, indicating that the geometry and connectivity of the block have been preserved.

Run # Initial highest reprojection error Final highest reprojection error Tie points removed
19 2.14 px RMS 0.99 px RMS 0.25%
20 2.14 px RMS 0.70 px RMS 1.21%

Key benefits

CONFIDENCE

Sub-pixel reprojection errors are an indicator of a consistent internal adjustment of the block, which raises confidence in metric analyses across the entire model.

INTEGRITY

Reprojection errors at sub-pixel level and balanced between images, together with a high redundancy of tie points, reduce the likelihood of geometrically weak regions, especially far from GCPs or in low-texture zones. A well-conditioned block in turn reduces the likelihood of noise, gaps and local distortions in dense point clouds, meshes and DEMs, improving their reliability and interpretability.

COHERENCE

The reduction of reprojection error at sub-pixel level with a limited loss of tie points, enables the refinement or extension of the orientation model, the update of GCPs and the co-registration of multi-epoch surveys on the same site to be more coherent.

STANDARDIZATION

Turning a manual sparse-cloud filtering stage to an optimization guided by the state of the block significantly reduces dependence on experiential choices and enables orientation results to be more comparable, reproducible and documentable.

Application scenarios

In light of the key benefits, MetricAware delivers its greatest value in scenarios where a high level of control over the robustness of the photogrammetric block is essential, namely:

  • cartography and architecture;
  • projects in which densification, meshing and digital elevation models support quantitative analyses (e.g. volumetric balances, hydraulic simulations);
  • multi-epoch monitoring on the same site (e.g. for deformation phenomena or seismic events);
  • surveys with control points that are limited in number/poorly distributed and/or characterized by complex geometries (e.g. oblique imagery, variable GSD, non-uniform coverage, low-texture zones), for which manual or excessively aggressive pruning of the sparse cloud can weaken the tie-point network.

Beyond purely automatic pruning

Originality and positioning

The joint review of the literature, commercial software and available open‑source solutions has not identified any tools that are fully comparable with MetricAware, which is the only solution that implements a per‑image adaptive optimization strategy. At each step, the iterative process identifies the image with the highest reprojection error and, within that view, selects the tie points to be removed, on the basis of local impact estimates, prioritizing those that contribute most to its reprojection error. This combination of per‑image targeting and local‑impact‑driven tie‑point selection enables a progressive refinement of the block which, as far as currently known, constitutes a specific prerogative of MetricAware in the landscape of automatic pruning technologies.

Keep in touch

CONTATTI

M +39 348 69 49 817
info@ileron.com

Ileron di Filippo Fiaschi
Headquarter Quarrata, Italy
Via Giovanni Falcone e Paolo Borsellino 40/C
P. IVA IT02376670465