Find the first incorrect spatial relationship

Find the first incorrect spatial relationship

Use this page when two or more independently reconstructed ASCAND scans should describe the same unchanged object, but Multi Merge does not produce a trustworthy shared result.

Keep these stages separate:

1. independent source reconstruction;
2. source selection and role assignment;
3. shared geometric overlap;
4. coarse alignment;
5. fine registration;
6. transformation acceptance;
7. merging into the shared reference; and
8. later meshing or export.

A problem at one stage can resemble a problem at another. Doubled surfaces can follow an incorrect transformation, residual registration error or an artifact already present in a source. A missing region can mean that the additional scan did not observe it, was not included, failed to align or contributed geometry that was later removed. A clean-looking seam can hide a globally wrong alignment on a symmetric object.

Begin with the accepted source reconstructions and move forward. Find the first stage at which their physical relationship becomes wrong.

Calculation completion, reported convergence and successful merging show that an operation produced a result. They do not prove that the transformation matches the physical object.

Review the general diagnostic path.

Preserve every source, role, transformation and result

Do not overwrite or clean the source scans before diagnosis. Keep:

• object and project identity;
• evidence that every scan represents the same physical object state;
• each original capture and independent reconstruction;
• Vision, Laser or Combo method for every source;
• accepted source point cloud or other documented input representation;
• reference and additional-scan assignments;
• order in which sources were introduced;
• preprocessing or filtering already applied;
• coarse initialization or documented manual input;
• fine-registration result and any available diagnostic message;
• every transformation associated with a registered source;
• intermediate registered derivatives;
• merged results after each addition;
• screenshots from the same global and overlap views; and
• every retry, re-export, repair or conversion already attempted.

Preserve the first failure and make separate working copies. Do not bake a transformation into the only remaining source file. Do not discard an earlier merged state when adding another scan.

This lineage matters because Multi Merge is incremental. If source B is incorrectly registered to reference A, then a later A+B result already contains the wrong spatial relationship. Adding source C may make the combined model denser while making diagnosis harder.

Record only metadata the platform or files actually expose. Do not reconstruct missing transformation values from appearance.

Follow the documented merge sequence.

Confirm that each source is independently usable and describes the same object

Registration should begin with accepted reconstructions, not with the hope that merging will repair them.

Inspect every source independently:

• Does it represent the intended object and project?
• Is the object rigid and unchanged between captures?
• Did a removable, articulated or flexible part move?
• Does the source contain a large reconstruction artifact?
• Is the expected new coverage actually present?
• Does the shared region contain usable geometry?
• Are the sources in compatible metric scale?
• Has the same scan accidentally been selected twice?

If a source is distorted, fragmented or missing the surfaces needed for alignment, correct its capture or reconstruction first. Use Vision Scan Problems or Laser and Combo Scan Problems as appropriate.

If the physical object changed, one rigid transformation may not describe the relationship. Registration can make one region agree while another remains displaced. That is not necessarily a fine-registration defect; it may be evidence that the source pair does not represent one unchanged rigid state.

Compatible scale is also a prerequisite. Multi Merge estimates the spatial pose of an additional reconstruction in the shared frame. Do not rescale a scan merely to force local correspondence. A scale discrepancy should be traced to source identity, project handling or another verified upstream transformation.

Test whether shared geometry can constrain one physical alignment

Complementary scans need two things: genuinely new observations and enough reliable overlap to determine how the scans relate.

Inspect the region visible in both sources. Strong overlap is not just a large count of nearby points. It must contain geometry that constrains the physical pose.

Registration becomes ambiguous when:

• only a small or narrow region is shared;
• the shared region is smooth or nearly planar;
• the object is rotationally symmetric;
• several features repeat with similar spacing;
• noise or outliers dominate the shared area;
• one reconstruction omits the distinctive features present in the other;
• point densities or boundary behavior differ substantially; or
• an occluding support or reconstruction artifact appears more consistently than the object.

A locally similar surface can support the wrong global placement. A cylinder, repeated tooth pattern or mirrored feature may allow several poses with comparable local separation.

Inspect the whole object, not only the apparent seam. Ask whether major features, ends, openings and asymmetric landmarks occupy the correct physical relationship.

There is no universal overlap percentage that proves registration quality. If the source pair contributes new coverage but cannot constrain a unique pose, plan another stable orientation that preserves more distinctive shared geometry.

Understand overlap and registration.

Separate coarse-pose failure from fine-registration misfit

Coarse alignment and fine registration solve different parts of the problem.

Coarse alignment establishes the approximate global pose. If it is wrong, the additional scan may be flipped, rotated to a repeated feature, placed at the wrong end of the object or remain separated by a large offset. Local refinement should not be expected to discover an entirely different physical relationship from an unsuitable initialization.

Fine registration refines already corresponding regions. ASCAND’s documented architecture includes Iterative Closest Point, or ICP, within this responsibility. Inspect for:

• a residual seam across shared surfaces;
• parallel or doubled surface layers;
• gradual local drift;
• one region fitting while another separates;
• a fit biased toward a large smooth area;
• correct local correspondence with the wrong global object pose; or
• an apparent improvement that removes no physical ambiguity.

ICP convergence means the optimizer reached a stopping state under its inputs and conditions. It is not independent proof that the physical alignment is correct.

If the global pose is wrong, return to the supported coarse-alignment or initialization step. If the pose is correct but overlap retains a systematic local separation, preserve the transformation and diagnose source quality, overlap and fine registration. Do not compensate by applying undocumented scaling, deformation or repeated random retries.

Learn how coarse and fine alignment differ.

Inspect transformation acceptance, merge order and every added contribution

Registration estimates a transformation. Merging applies accepted transformed geometry to the growing reference result. These are related but distinct actions.

Before merging, compare:

• the untouched reference;
• the untouched additional source;
• the additional source transformed into the reference frame; and
• the shared overlap at both global and close views.

After merging, confirm that the intended contribution remains identifiable. Look for doubled surfaces, implausible thickness, intersecting parts, unexpected tilt, retained support artifacts, new outliers, a missing contribution or a seam that appears only after combination.

Inspect each addition before continuing. A clean A+B result can establish a stronger reference for C. An incorrect B transformation can instead contaminate the cumulative result and influence later registration.

More points do not necessarily mean more correct evidence. The same incorrect surface may be duplicated, and a dense overlap can conceal the fact that a thin feature has become two parallel layers.

Keep the merged point cloud distinct from a later mesh. Surface reconstruction may bridge, smooth or close regions and can hide a registration seam without correcting the underlying spatial relationship. If the registered cloud is sound and the first defect appears during meshing or export, use Mesh and Export Problems.

Change one registration variable and compare the same overlap

Use a controlled comparison only when current documentation supports the relevant action.

Keep constant:

• the original source reconstructions;
• the chosen reference;
• the additional scan;
• the inspected overlap regions;
• the global viewpoints; and
• the intended physical relationship.

Change one documented factor, such as the supported initialization or whether one questionable additional source is included. Preserve both results and record the change.

Evaluate more than visual closeness:

1. Is the global pose physically correct?
2. Do distinctive shared features correspond?
3. Are seams or doubled layers reduced throughout the overlap?
4. Does the new coverage appear in the correct place?
5. Did another region become worse?
6. Does the transformation remain consistent with a rigid unchanged object?

A lower reported residual, if one is exposed, is evidence about the optimizer’s chosen comparison. It is not sufficient by itself. A symmetric but wrong placement can score well locally.

Do not run many undocumented combinations and keep only the most attractive result. That destroys the causal value of the comparison and can select a plausible-looking false alignment.

Evaluate the merged evidence.

Correct the earliest confirmed fault or escalate with reproducible evidence

Choose the route that owns the first confirmed problem:

• The fault is visible in the capture: use Capture Setup Problems.
• One independent reconstruction is unsuitable: use the relevant Vision or Laser/Combo troubleshooting page.
• The sources do not share enough distinctive geometry: plan a complementary recapture that preserves new coverage and overlap.
• Current controls, accepted inputs or interface states are unclear: use Documentation.
• The sources are accepted but the physical relationship becomes wrong during registration: preserve the complete case and contact Support.
• The registered point cloud is sound and the first fault appears in a mesh, export or import: use Mesh and Export Problems.

For escalation, include the smallest privacy-cleared set that reproduces the problem: project and source identities, methods, unchanged-object statement, reference and additional roles, source order, accepted source results, transformations or available registration records, intermediate and final results, exact messages, and annotated global and overlap views. State which stage first becomes wrong and what one-variable comparisons were attempted.

Do not share unrelated project data or credentials.

A successful retry, accepted transformation or visually clean merge can support continued evaluation. It does not certify accuracy, completeness, dimensional suitability or fitness for a downstream use.

Contact Support.Use this page when an identifiable ASCAND result exists, but a generated mesh, export, downloaded file or imported model behaves unexpectedly.