1. Observe
The camera receives existing light or light from its own projector or laser.
Knowledge · 3D cameras
An industrial 3D camera measures the distance to visible points on a product or in a work area. The resulting depth map or point cloud allows a vision system to determine height, shape, volume, position and orientation for measurement, inspection and robot guidance.

The basics
A normal 2D image describes where light, dark or coloured pixels are located. A 3D camera adds distance information for visible surfaces. After calibration, software can convert that data into height, dimensions, volume, or the position and orientation of a part.
The camera does not see hidden internal geometry. It measures only surfaces that return enough light from the selected viewing direction and can be observed by the measurement principle being used.
How a 3D camera creates depth
The underlying physics differs between cameras, but the processing chain is similar.

The camera receives existing light or light from its own projector or laser.
The system derives distance from image disparity, pattern deformation, travel time or triangulation.
The measurement becomes a depth map, point cloud or height profile.
Software calculates a dimension, checks a limit or guides a robot.
Output
An image grid with a distance value at each valid pixel. For an illustrative pixel at row 120, column 80, a stored value might represent 750 mm. Check the data scale and whether distance means range along a viewing ray or depth along the camera axis. Display colours usually indicate a selected distance scale, not the product’s actual colour.
A set of X, Y and Z coordinates representing visible surface points. A hypothetical point (25, −10, 750) mm means nothing without the coordinate origin and axis directions. Camera calibration can convert valid depth pixels to spatial points; a further transformation relates them to the machine or robot.
A cross-section through the surface. A moving product or sensor can build successive profiles into a height image. The distance between profiles depends on measured movement and acquisition timing, so movement errors can distort the resulting geometry.
Colour registered to depth so software can associate appearance with a measured surface. RGB-D describes the combined data, not a separate measurement principle. Confirm registration and timing when colour and depth come from different sensors, especially on moving objects.
Applications
3D adds value when the process decision depends on height, shape, volume or a position outside one fixed plane. For presence, colour, codes or outlines in a plane, 2D may be simpler.

Dark, glossy or transparent surfaces, strong ambient light, motion and hidden regions can produce missing or incorrect depth. An invalid value is not evidence of an empty location. Preserve validity information and test how the application handles holes, edge artefacts and multiple reflections.
To estimate object height above a table, establish the table plane and measure the surface relative to it. Camera distance alone is not object height, particularly with an angled view. Validate the resulting height or position against an independent reference throughout the required volume.
Products
The product category helps you compare available industrial 3D cameras. Use it after establishing which data and working volume the application needs.
Measurement principles
Determines distance from the difference between two images, optionally with projected texture.
Measures how a known light pattern deforms across the surface.
Derives distance from the travel time or phase of emitted light.
Measures the position of a laser line and builds a profile or height image during motion.

Practical examples
Container or pallet:a point cloud helps a robot find a reachable part or free position.
Conveyor:a height image reveals volume, deformation or missing geometry.
Weld or sealant bead:a profile makes width, height and interruptions measurable.
The camera supplies measurement data. Vision software removes invalid points, relates the data to a coordinate system and then calculates the outcome required by the process.
That outcome may be a dimension, pass or fail result, volume estimate or robot position. The PLC or robot uses this outcome and does not always need the complete point cloud.
Practical trial
Evaluate the complete measurement chain using representative parts and predefined limits. Acceptance concerns the process result, not only a visually clean point cloud.
Test accepted, rejected and borderline cases from the actual product mix. Include relevant surfaces, colours, shapes and orientations.
Vary measurement distance, part position, motion, ambient light and the contamination that may occur during normal production.
Define where valid depth points are required and how many missing or incorrect points the calculation may contain.
Test a reference object with known dimensions at near and far distances and across the field. Repeat with representative surfaces and motion. Record bias, repeatability and missing-data fraction separately; a dense point cloud alone does not establish measurement uncertainty.
Measure capture, transfer, processing and communication together at the highest agreed line load. Include any motion or multiple captures.
One capture cannot establish suitability for every material or orientation. Define the valid measurement range and the outcome when required data is absent. Include the reference test after calibration or hardware replacement so maintenance can verify restoration.
Business case
A 3D camera creates process value only when its measurement result demonstrably affects manual work, errors or downtime. Use data from your own line and record assumptions separately.
Measure manual measurement or inspection time, operator interventions, staffing per shift and time spent recording data.
Collect quantities and costs for rejects, material loss, rework, returns, claims and line stops that the 3D measurement can influence.
Record volumes, cycle time, batch size, product mix, changeover time and available maintenance windows.
Include the sensor, controller, mechanics, shielding, calibration, software, integration, validation, training, cleaning, spares and licences.
Annual net benefit = avoided manual hours × applicable hourly cost + avoided error and downtime costs + contribution from usable added capacity − added annual operating costs. Express all terms in money over the same period and avoid double counting. Compare this with total investment, and state which inputs still require measurement in a trial.
No. Some systems provide only distance or intensity. RGB-D systems register depth to colour.
Not from one viewing direction. Hidden surfaces require another position, multiple cameras, or movement of the sensor or product.
No. 3D adds depth, but achievable measurement quality depends on the principle, geometry, surface and conditions.
When contrast, colour, a code or an outline in one known plane fully answers the process question.
Define the height, shape or position required by the process and bring representative parts to a trial. For the investment comparison, include calibration, software, integration and validation hours × rate alongside hardware. Compare them with measured manual measurement or handling work; depth data alone does not establish a saving.