Working volume and mounting
- Minimum and maximum working distance
- Required field of view in width and height
- Depth variation within the object or container
- Available mounting angle, space and dead zones
Knowledge · 3D camera selection
How do you select a 3D camera for an industrial application?
A useful shortlist does not start with camera models but with the decision the system must make. Define the working volume, tolerance, surface, motion and integration first, then test only candidates that can meet those constraints.
Step 1 · The application
Describe the outcome the process needs first
The same depth map can be more than adequate for a presence check and unsuitable for dimensional inspection. Do not merely state that you need “3D data”; document the process decision it supports and the error that cannot be accepted.
Detect
Is a part present, complete or within a height limit?
Measure
Which height, shape, volume or dimension must be determined, and to what tolerance?
Locate
Which position and orientation does a robot or downstream process require?
Inspect
Which deviation must be visible and what happens after a pass or fail?
Check whether 3D is necessary
If colour, outline or position in one plane answers the question completely, a 2D setup may be simpler to integrate and maintain. Choose 3D when height, depth, volume or variable orientation is genuinely part of the decision.
Compare 2D and 3D robot vision →Step 2 · The requirements
Make every selection requirement measurable
Avoid words such as fast, accurate and large. Record limits, conditions and an acceptance method. This lets suppliers answer the same question and allows candidates to be compared under equal conditions.
Measurement performance
- Smallest relevant feature or height change
- Permitted error in the final process result
- Required repeatability and, where needed, absolute accuracy
- How much missing or outlying data the algorithm can tolerate
Part and environment
- Colour, gloss, transparency and surface texture
- Variants, contamination and permitted part movement
- Ambient light, dust, vibration and temperature
- Potential interference between active 3D sensors
Process and integration
- Available capture time and maximum decision latency
- Trigger, encoder, robot or PLC connection
- Required output: depth map, point cloud, profile or direct measurement result
- Calibration, diagnostics, logging and replacement procedure
Step 3 · Geometry
Check range and sampling before comparing models
A camera must see the complete working volume from the planned mounting distance. Divide the field of view by the number of measurement columns for an initial estimate of lateral sampling distance. That number is not accuracy: optics, triangulation geometry, calibration, noise and the surface determine how much measurement information is actually usable.
Check performance at the edges and at the minimum and maximum distance. A specification at the centre of the range does not automatically describe the complete working volume.
Keep four terms separate
Sampling
The spatial distance between measurement points.
Repeatability
How closely repeated measurements agree under equal conditions.
Absolute accuracy
How close a measurement is to a traceable reference value.
Resolution
A term that says too little for selection without a direction, test condition and definition.
Step 4 · Measurement principle and shortlist
Only then choose an appropriate technology direction
Passive and active stereo, structured light, time of flight and laser triangulation respond differently to surface, motion, range and ambient light. Use the requirements to eliminate unsuitable directions. RGB-D and point clouds are output representations, not independent measurement principles.
Compare the 3D measurement methods →Eliminate
Remove candidates that demonstrably cannot handle the working volume, motion or critical surfaces.
Compare
Compare the remaining models against the same requirements, interfaces and total hardware needs.
Test
Select a small shortlist for measurements on real parts under representative conditions.
Step 5 · The practical test
Define in advance when a candidate passes
Do not test only the easiest part. Use the relevant variants, extreme distances, normal motion, actual ambient light and planned processing time. Then evaluate the output of the full algorithm, not only how attractive the point cloud looks.
Coverage
Do critical surfaces and edges remain measurable for every variant?
Measurement quality
Do error, repeatability and outlier count meet the predefined limit?
Process result
Does the robot, inspection system or controller demonstrably make the correct decision?
Time and recovery
Do capture, processing and any repetition fit within the available time budget?
Step 6 · Total cost
Compare the complete solution, not only the camera
In addition to the sensor, include the projector or illumination, mounting, cabling, industrial computer, software licences, calibration, integration and maintenance. A camera with a lower purchase price can cost more when it requires additional shielding, processing capacity or frequent intervention.
Connect the choice to the business problem: less manual inspection or handling, less downtime, higher capacity, better ergonomics or more consistent quality. Use only effects that can be measured during the trial or in the existing process.
Ready for product screening?
A shortlist becomes useful once the measurement task, working volume, critical parts, timing and integration requirements have been documented.