How to create a 2D image from laser triangulation

A laser-triangulation scan can support blob and edge analysis, but first decide what each image pixel should represent. A height image describes surface geometry; a laser-return image describes the detected laser signal. Neither automatically provides the appearance you would obtain with a separately illuminated camera.

For long parts, the practical solution usually combines calibrated motion with a defined output grid. This lets you assemble profiles into a continuous height image, extract geometric features, and decide whether an additional reflectance view is necessary. The key is preserving physical coordinates and missing-data information throughout that process.

Decide which kind of 2D image you need

A two-dimensional array is an image format, not a guarantee of ordinary grayscale appearance. Several arrays from the same triangulation acquisition can represent very different quantities.

Choose height when the defect is geometric. Choose a separate camera view when the feature is primarily ink, discoloration or surface texture. A laser-return channel may help, but evaluate it on representative parts before treating it as reflectance.

Also distinguish the raw triangulation camera frame from the reconstructed surface. The raw frame typically contains a displaced laser stripe, not a complete overhead view of the part.

Check whether your height map is already an image

An organized height map may already be the required input for blob analysis. If its columns represent calibrated cross-track positions and its rows represent calibrated travel positions, each valid pixel contains a height value.

You do not need to convert that map into a point cloud and back merely to threshold it. Instead, retain the original height values, establish a valid-data mask, and apply the operation to a suitable height range or residual surface.

However, an organized array is not necessarily a uniformly spaced Cartesian grid. Check the acquisition documentation for:

• The meaning, units and orientation of each axis.
• Whether cross-track spacing is uniform after calibration.
• How motion determines the position of each acquired profile.
• The representation of invalid samples and missing profiles.
• Whether multiple output channels share identical sampling coordinates.

Coordinate channels such as X, Y and Z describe location. They are not three alternative appearance images. Likewise, a score channel should not be interpreted as brightness unless the documented definition supports that use.

If the extracted stripe itself is unstable, address acquisition quality before rasterization. The guide to reducing laser-line extraction noise explains why unreliable stripe positions propagate into unreliable geometry.

Rasterize calibrated points onto a physical grid

When the input is an unorganized point cloud or a nonuniform coordinate map, create a top-down height raster in a chosen reference frame. Define the grid spacing from the smallest feature you need to inspect, not from the display window size.

For the following equations, X₀ and Y₀ are the lower boundaries of the raster. Column and row indices start at zero.

Here, ΔX and ΔY are physical sample spacings. Transform the points into the inspection reference frame before assigning them to cells. Otherwise, a tilted reference surface can appear as an unwanted height gradient. Use the triangulation tilt and calibration checks when the slope may come from alignment or calibration.

A practical rasterization sequence is:

1. Reject invalid coordinates and samples that fail documented quality checks.
2. Transform valid points into the inspection reference frame.
3. Assign each point to a raster cell.
4. Resolve cells containing multiple points using a defined rule.
5. Mark cells with no observation as invalid.
6. Retain a coverage or sample-count map for verification.

The multiple-point rule changes the result. Averaging can suppress noise but soften narrow peaks. A maximum-height rule preserves the highest observed surface, but can also preserve positive outliers. Select the rule around the inspection objective and validate its bias.

A height raster represents one elevation per cell. It cannot preserve every surface of an overhang or stacked geometry. Keep the point cloud when the task needs multiple surfaces along the same viewing direction.

Turn height into useful analysis contrast

For geometric inspection, the most useful image is often a height residual rather than absolute elevation. Subtract a reference plane or reference surface so that expected part position contributes less contrast than the defect.

Fit the reference only to suitable regions. Including a large dent in the fitting region can move the estimated reference and reduce the apparent defect depth.

For example, display residuals between −0.50 mm and +0.50 mm using an 8-bit image. Zero residual maps to approximately gray level 128. The nominal display increment is:

That increment describes display quantization, not measurement accuracy. Noise, calibration and optical geometry still govern the trustworthy height information.

Retain floating-point or original calibrated height data for measurement. An inspection threshold of +0.20 mm should be applied to physical residual height, rather than inferred from a screenshot or a color palette.

Use a fixed display range across comparable parts. Automatic per-image normalization can make a small deviation look as prominent as a large one.

Extract blobs and edges without changing the measurement

For a raised-feature inspection, threshold valid residual heights, label connected regions, and filter them by physical area or dimensions. Keep the validity mask separate from the thresholded feature mask.

With uniform sampling, physical blob area follows directly from pixel count:

For example, a region containing 240 pixels on a grid with 0.10 mm cross-track spacing and 0.20 mm travel spacing has an estimated area of:

This is projected area on the raster plane, not the true area of a sloped surface. Boundary sampling and missing data also affect the estimate.

For dents on an inclined part, use local reference geometry rather than a single absolute-height threshold. The guide to surface blob analysis for dents on tilted parts develops that distinction further.

Edges require an equally clear definition. A height step, an appearance boundary and the boundary of valid observations are different things. A missing laser return must not automatically count as a physical part edge.

If you compute gradients, divide by the corresponding physical spacing. Unequal axis spacing otherwise biases edge strength and orientation. Any smoothing neighborhood should also have a defined physical size.

For a long, consistent feature, averaging several neighboring profiles can stabilize a measurement signal. The guide to building a stable intensity profile by summing image rows explains profile aggregation; apply the same missing-data discipline when the input represents height. For dimensional acceptance, distinguish sampling from accuracy using the edge-position accuracy guide.

Build long-part coverage from motion, not image stretching

A narrow instantaneous field of view does not prevent inspection of a long part. Repeated profiles can cover its length, provided their physical positions are known.

Consider an illustrative inspection area measuring 80 mm across the scan and 1,200 mm along travel. Choose output spacings of 0.10 mm across and 0.20 mm along travel:

These are requested output dimensions, not evidence that the sensor resolves every cell independently. Interpolation cannot create detail absent from the acquired measurements.

At a travel speed of 400 mm/s, the required profile rate for that along-track spacing is:

The same distance-per-line relationship applies when planning a separate line scan view.

Use measured travel position when speed varies. A fixed time interval produces unequal physical spacing during acceleration or speed changes. Check the acquisition system's documented position-triggering or timestamp-and-position capabilities rather than assuming them.

The closest companion guide, preventing stretched triangulation point clouds through correct scan spacing, explains the geometry behind this problem. For a separate line scan camera, follow the line-rate setup procedure.

For multiple calibrated views, transform observations into the same reference frame before combining them. Overlap should support registration checks, not simply hide a seam. Repeated visual texture alone is often an ambiguous registration target.

Add a separate appearance view when height is insufficient

Use a separate camera when the inspection depends on print, color or texture that does not produce dependable geometric contrast. This also separates the appearance-lighting problem from the triangulation reconstruction problem.

A line scan view suits continuous travel and can share the part-position reference with the height acquisition. An area scan view suits complete-part coverage or a sequence of calibrated image tiles.

The diagram shows the separate appearance-imaging arrangement, not the triangulation reconstruction geometry. Compare line scan cameras with area scan cameras in the camera finder.

LS-LCM-0080 is a monochrome GigE line scan option with an M42 mount. MV-CU020-90GC is a color GigE area scan option with a global shutter. These are candidates for the separate appearance channel, not substitutes for a calibrated triangulation sensor.

Select optics around coverage and required object sampling. The guides to calculating line scan coverage and preserving features across acquisition frame boundaries address separate parts of that planning task.

For continuous appearance capture, line scan lights concentrate illumination near the observed line. Use the lighting finder to compare candidates. LS-LSC-0527 is a line light source to evaluate against the required illuminated width and contrast geometry.

Electrical compatibility, useful illumination at the part, exposure requirements and control interfaces still need verification. The guide to lighting a fast inspection line explains why nominal light dimensions alone are insufficient.

Register the appearance image and height data using common physical coordinates. A simple planar image transform may be inadequate when perspective and changing part height produce parallax. Validate registration across the actual height range and throughout the inspection area.

Know when the usual conversion is wrong

Rasterization cannot recover a surface the triangulation geometry never observed. Shadows, steep walls and weak returns can leave genuine gaps. Filling those gaps may improve display appearance while making inspection evidence less trustworthy.

Keep interpolation conservative. A small missing region on a smooth reference surface is different from a gap beside a suspected crack or step. Store the original validity mask even if you produce a filled image for visualization.

A separate camera does not guarantee complete coverage either. Its viewing direction, illumination and part geometry still determine visibility. Multiple views may be necessary when the task includes sidewalls or hidden features.

Finally, avoid converting everything to grayscale merely because an existing tool expects it. If the software supports calibrated height directly, preserve those units. If it accepts only a limited image type, document the scaling, clipping and invalid-value handling as part of the inspection configuration.

Checks before specifying the system

Use this checklist before selecting the appearance camera or finalizing the processing pipeline:

• Define the feature: state whether acceptance depends on height, appearance or both.
• Document channel meanings: confirm units, coordinate frames, validity codes and quality-channel definitions.
• Set physical sampling: specify cross-track and travel spacing independently, with a feature-based justification.
• Verify motion coverage: check profile positions, missing acquisitions and start-to-end dimensional consistency.
• Preserve measurement data: retain calibrated heights and masks alongside any display image.
• Test difficult geometry: include steps, edges, slopes and regions with missing returns.
• Validate combined views: check registration across the full length and expected height range.
• Check interfaces and illumination: confirm acquisition support, triggering, exposure and lighting compatibility before ordering.

Start with a representative scan and a written feature definition. If height alone answers the inspection question, keep the pipeline geometric. If appearance is also required, use the camera configurator to plan that channel, or contact the applications team with coverage, sampling, motion and example-data requirements.