Knowledge · 3D sensing
Passive and active stereo, structured light, time of flight and laser triangulation measure depth in different ways. The right choice therefore starts with the measurement task, the part and the production environment, not a single specification.
Set the terminology first
A measurement principle describes how a sensor determines distance. RGB-D describes a colour image with a spatially registered depth map. That RGB-D output can be produced by active stereo, structured light or time of flight, among other methods.
A point cloud is also a data representation: depth values have been converted into 3D coordinates. The format alone says nothing about the measurement method, accuracy or suitability for an industrial application.
Five widely used principles
The properties below are directional. Optics, baseline, wavelength, illumination, algorithm and implementation vary by camera. Always test with representative parts.
01 · Triangulation without a projector
Two or more cameras find corresponding image points from different viewpoints. The displacement between those images produces depth through triangulation.
Often considered for: scenes with sufficient natural texture and applications where a pattern cannot be projected onto the object.
Watch for: Correspondence can be difficult on uniform surfaces, with repeating patterns, in low light and where parts partially hide one another.
02 · Stereo with added texture
A projector adds a recognisable, often random pattern. The cameras still calculate depth from the difference between their viewpoints; the pattern helps them find corresponding points.
Often considered for: objects with little natural texture and full 3D captures within a defined working volume.
Watch for: Strong ambient light, gloss and multiple projectors can interfere with the added pattern.
03 · Deformation of a known pattern
A projector casts a known pattern onto the object. One or more cameras measure its deformation and derive the geometry. Some systems use one exposure, while others project a sequence of patterns.
Often considered for: detailed shape measurement on stationary or controlled moving objects.
Watch for: Motion during a pattern sequence, direct sunlight and highly reflective or transparent surfaces can produce gaps or errors.
04 · Light travel time
The sensor emits modulated light and determines the distance per pixel from the time or phase difference of the returning radiation. This creates a depth image of the full field in one capture.
Often considered for: larger working volumes, fast full-field depth captures and positioning where very fine surface detail is not the primary requirement.
Watch for: Multipath reflections, edges, strong ambient light and low reflectivity can influence distance measurements.
05 · Line profile by triangulation
A laser projects a line onto the surface. A camera observes the displacement of that line from a known angle and calculates one height profile. Controlled movement of the part or sensor combines consecutive profiles into a 3D surface.
Often considered for: conveyor inspection, height and profile measurement, and applications where controlled motion is available.
Watch for: speed, encoder information and motion stability determine scale along the direction of travel. It is not an immediate full-field capture of a stationary scene.
Selection criteria
Presence, robot positioning, volume and dimensional inspection do not require the same point density, repeatability or absolute accuracy.
Define minimum and maximum distance, field of view, dead zones and the space required for camera and projector.
Test black, glossy, metallic, transparent and variable parts separately. Average samples often hide the most difficult variant.
Determine whether the object stops, how much capture time is available and whether multiple exposures or line scanning fit the process.
Ambient light, dust, vibration, temperature and other active sensors can alter measurement quality and availability.
Include calibration, mounting stability, processing load, interface, diagnostics and replaceability in the architecture choice.
Evidence for the investment
A feasibility test should approximate the parts, distances, mounting angle, motion and illumination of the final situation. Do not judge only an attractive depth image; measure how much usable data remains available on critical surfaces.
Record the same metrics for every candidate: coverage, repeatability, deviation from reference dimensions, outlier count, processing time and behaviour after contamination or displacement. This prevents an inexpensive sensor from causing extra manual work, repeat captures or machine downtime later.
Continue reading
Next step
The best candidate is the method that provides enough usable measurement data for a correct decision under your production conditions. A focused trial reveals that boundary before you invest in the complete installation.
