Multi-Scan and Multi Merge
Multi-Scan captures the same physical object in more than one orientation. Multi Merge then registers and combines the independently reconstructed geometry.
The distinction matters because one turntable rotation does not necessarily expose every surface. The area touching the turntable remains hidden. The object may block the camera’s view of an underside, recess or overhang. A new physical orientation can reveal some of those regions.
Each orientation is still a separate scan. It has its own capture observations, reconstruction and local coordinate frame. Before the geometry can form one result, processing must estimate how the additional point cloud is rotated and translated relative to a chosen reference.
That operation is registration. Merging comes afterward.
Additional orientations can extend observed coverage, but they do not guarantee a complete or correct model. The scans must describe the same unchanged object, retain compatible scale, contain enough trustworthy overlap and align to the correct geometric relationship. Regions unseen in every orientation remain unobserved.
Follow the verified Multi Merge workflow
A full rotation does not make every surface visible
ASCAND changes the relative view by rotating the object in front of a steady camera. This creates observations around the rotation axis, but rotational coverage is not the same as complete surface coverage.
Several regions may remain unavailable:
- the surface resting on the turntable;
- an underside blocked by the body of the object;
- a deep recess whose opening never faces the camera;
- one side of an undercut or overhang;
- geometry hidden behind another part of the object;
- an optically difficult region that does not produce usable evidence.
Placing the same object on its side changes which surfaces face the camera and which surface touches the turntable. The second scan may therefore observe regions that were hidden in the upright scan.
The change also creates a planning trade-off. If the second orientation reveals only new geometry and preserves almost no recognizable part of the first scan, registration may lack the shared evidence needed to align the two results. Useful Multi-Scan capture needs:
- complementary visibility, so an additional scan contributes something new; and
- reliable overlap, so the separate reconstructions can be placed into a common coordinate system.
More orientations are not automatically better. Each scan adds observations, but it also adds another reconstruction, transformation and potential source of noise or misalignment.
Understand what rotation contributes
Each capture becomes an independent reconstruction
Multi-Scan does not combine raw views from different object placements as though they belonged to one uninterrupted rotation.
Each physical orientation is captured and reconstructed independently. Depending on the selected method, the result may contain Vision-derived geometry, laser-derived samples or a combined representation. Multi Merge receives processed geometric results—principally point clouds—rather than replacing the responsibilities of acquisition and reconstruction.
The representations should remain distinct:
| Stage | What it represents |
|---|---|
| Capture observations | Images or video evidence obtained in one object orientation |
| Independent reconstruction | Geometry estimated from one capture sequence |
| Local point cloud | Surface samples expressed in that scan’s coordinate frame |
| Transformation | Estimated rotation and translation from one coordinate frame to another |
| Merged point cloud | Registered geometry accumulated in a shared reference frame |
| Mesh | A later connected surface constructed from geometric evidence |
Two point clouds can have the correct physical scale and still occupy different coordinate frames. In one scan the object may stand upright; in another it may lie on its side. Their coordinates are therefore not directly comparable until the pose difference is estimated.
A file containing both unregistered clouds is not a merged object. It is only two datasets occupying one file or scene.
Understand point clouds and meshes
Shared geometry makes registration possible
Registration estimates the spatial transformation that maps an additional point cloud into the reference coordinate system.
For a rigid, unchanged object, that transformation consists of rotation and translation. Scale should already be compatible. Registration seeks geometric correspondence: regions in one cloud that plausibly represent the same physical surfaces as regions in the other.
Overlap provides the evidence for that correspondence. Broad, distinctive shared surfaces and features can constrain the transformation. Weak or ambiguous overlap makes the problem harder.
Registration may be challenged by:
- too little shared geometry;
- highly symmetric or repeated shapes;
- smooth forms with few distinctive features;
- different point densities or missing regions;
- outliers, noise and reconstruction artifacts;
- large initial pose differences;
- flexible, articulated or moved parts;
- an object that changed between captures.
Symmetry deserves special attention. Several alignments may look numerically plausible when the object repeats around an axis or contains similar parts. An algorithm may reduce local point-to-point distance while placing the scan in the wrong rotational state.
This is why “the calculation converged” and “the geometry is correctly aligned” are not equivalent statements. Registration produces an estimate that must be evaluated against overlap, object structure and the intended use.
Diagnose Multi Merge registration problems
From complementary captures to a merged point cloud
A public conceptual model contains seven stages.
1. Plan complementary orientations
Identify surfaces hidden in the first placement and choose another stable placement that reveals useful geometry while retaining recognizable overlap. Confirm that the object may be safely repositioned and will remain unchanged.
2. Capture each scan
Record each orientation as its own valid scan under the requirements of the selected Vision, Laser or Combo method.
3. Reconstruct independently
Process each capture separately. Inspect the individual results before merging; a merge should not be expected to repair a fundamentally unusable source scan.
4. Prepare the point clouds
Applicable preparation may remove unrelated support-plane geometry, reduce outliers, normalize data for registration or create working-resolution representations. Preparation should preserve the evidence needed for alignment.
5. Establish coarse alignment
Estimate the large pose difference between the additional cloud and the reference. This stage brings corresponding regions close enough for local refinement.
6. Refine and assess alignment
Fine registration reduces the remaining separation between overlapping surfaces. The transformation and overlap must then be inspected for seams, duplication, false correspondence and unsupported placement.
7. Merge into the shared reference and inspect
Transform accepted geometry into the reference coordinate system and add its supported contribution. The merged point cloud becomes input for review, further scans, export or later surface reconstruction.
The sequence separates acquisition, reconstruction, registration and merging. Current interface controls, preprocessing options and available outputs belong in the verified tutorial and platform documentation.
Review and download scan results
Registration solves large pose differences before refining local fit
When two scans begin in very different poses, fine local alignment alone may not know which surfaces correspond.
Coarse or global registration addresses the large pose difference. It searches for a plausible initial transformation using geometric structure, features or orientation hypotheses. Its purpose is to bring the clouds into approximately the same pose.
Fine registration then refines that estimate using nearby geometric correspondence. ASCAND’s documented Multi Merge pipeline uses Iterative Closest Point, or ICP, as part of this responsibility. Conceptually, ICP repeats a cycle:
- identify likely corresponding points or surfaces;
- estimate a transformation that reduces their separation;
- apply the transformation;
- repeat until the improvement stabilizes or another stopping condition is reached.
ICP is an optimizer, not a proof system. If the initial pose is wrong, the overlap is weak or the object is symmetric, it can settle into a locally plausible but physically incorrect alignment. Noise and unequal coverage can also bias correspondence.
The transformation remains valuable because it makes the spatial decision explicit. But its quality must be judged through the registered geometry, not through the existence of a transformation matrix alone.
A merge preserves evidence; it does not automatically resolve every disagreement
After registration, the additional scan is transformed into the reference coordinate system. Its geometry can then extend the shared result.
Multi Merge is cumulative: a primary scan establishes the reference, an additional scan is registered and merged, and the growing result can become the reference for another scan. The order therefore has meaning. A weak early alignment can affect later comparisons.
Overlapping scans may produce several outcomes.
Consistent overlap. Shared surfaces occupy compatible positions. The additional scan can extend coverage without creating a visible duplicate.
Residual seam or doubling. Small offsets leave parallel surfaces, fuzzy boundaries or discontinuities. Causes can include imperfect registration, different reconstruction behavior, noise or deformation.
False alignment. The optimizer matches the wrong repeated or symmetric region. The numerical fit may appear acceptable locally while the overall object is wrong.
Complementary gap filling. A later scan supplies observed geometry where the reference has no evidence. This is the intended coverage benefit, provided the transformation is correct.
Shared missing region. No scan observed the surface. Merging cannot turn that absence into measured geometry.
Implementation rules for overlapping points, colors, normals, outlier removal and retained outputs can change with the platform version. They should be described in current documentation rather than converted into timeless quality promises.
Successful processing also does not mean the result is watertight, dimensionally certified or ready for 3D printing. Meshing, hole handling, smoothing and other downstream operations have separate responsibilities and can introduce inferred geometry.
Understand output formats and post-processing
Evaluate the merged evidence before building the final surface
A Multi Merge result should be inspected as geometric evidence.
Check:
- Object identity: Every source scan represents the same unchanged physical state.
- Scale: The reconstructions use compatible metric scale.
- Overlap: Shared regions are broad and distinctive enough to support the transformation.
- Global pose: Major features occupy the physically correct relationship.
- Seams and duplication: Overlap does not create doubled surfaces or implausible thickness.
- New coverage: Additional geometry corresponds to a surface genuinely observed in the complementary scan.
- Remaining gaps: Unseen regions remain identified rather than being mistaken for captured geometry.
- Downstream suitability: The merged cloud supports the intended evaluation, modeling, fabrication or presentation task.
Manual refinement can be useful when automatic registration is close but not sufficient. It does not remove the need for shared evidence or turn an ambiguous object into a uniquely solvable alignment problem.
Do not confuse Multi-Scan with Combo scanning. Combo uses different evidence types—Vision and Laser—within a related acquisition workflow. Multi-Scan uses separate object placements and independently reconstructed geometry. A Combo result can itself become one input to a Multi Merge workflow, but the two concepts retain different responsibilities.
Use the current tutorial for operation and troubleshooting for failed or implausible alignment. Use the representation and limitations articles to judge what the merged point cloud means before meshing or export.
Diagnose Multi Merge problemsCombo scanning uses complementary visual and laser observations of the same object within one coordinated reconstruction workflow.