Stereo vision cameras for robots: baseline, texture, and depth

A stereo vision camera estimates depth by relating observations from different viewpoints. For a robot buyer, the practical question is whether the required surfaces produce dependable depth across the working volume. Baseline, texture, lighting, and occlusion should be evaluated together with the intended task.

Understand the correspondence problem

IDS projected texture stereo principles explains that stereo matching depends on corresponding features in two images. A surface visible in only one view cannot be treated like a well-observed shared feature. Ask the supplier to identify where your container walls, deep recesses, or adjacent parts reduce usable coverage.

Include samples with the actual texture and finish. Smooth plastic, repetitive patterns, black surfaces, and reflective metal should be present if they belong to the supported product family. If the proposed system projects texture, test it in the installed lighting and with other active sensors operating as planned.

Do not rank cameras by baseline alone. Ask for evidence at the closest and farthest required distances, with the feature size your task uses. A physically wider unit may also change where it can be mounted and which views remain available. The 3D camera checklist helps organize this comparison.

Draw shared visibility into the layout

Have the vendor show the useful observation region in the cell model. Include the robot and gripper, not just the bin. A camera can have a suitable distance range while its view becomes blocked during the capture part of the sequence.

For fixed cameras, consider whether different bin placements remain observable. For wrist-mounted cameras, specify the capture poses and cable path. Use our camera mounting comparison to evaluate whether moving the viewpoint is worth the added mechanical and timing responsibilities.

Verify calibration and output

OpenCV camera calibration tutorial documents camera calibration and distortion correction. Ask how the stereo system’s calibration is maintained, which installation changes require verification, and what diagnostic data your team can access. Factory calibration and robot-coordinate calibration should be identified as separate responsibilities where both apply.

Require the output convention in writing: units, axis directions, valid-data flags, and the transformation to the robot frame. Include an independent check of known positions across the required volume. A depth image that looks convincing is not enough to establish a correct robot target.

Compare depth at production timing

Hypothetical example: one setup improves surface coverage by combining multiple captures of a stationary scene. Another provides less dense output quickly. If the process can pause, the first may be suitable; if parts keep moving, the evaluation must show that the chosen acquisition method remains valid. Treat this as a task decision, not a universal ranking.

Measure capture-to-usable-result time, including transfer and processing on the intended computer. Record the rate of valid targets and the reason for rejected results. For bins, test until the agreed empty condition is reached using the bin-picking acceptance procedure.

Request comparable evidence

EMVA 1288 camera characterization standard standardizes camera characterization. Use component data where relevant, but require separate evidence for the stereo system’s depth output and your final robot operation. Ask whether quoted performance applies across the whole working range or only under one favorable test condition.

The handover should contain saved sample captures, settings, calibration records, installation drawings, and recovery instructions. Specify how a replacement camera is introduced and verified. A supported operating envelope and a reproducible test are more useful purchasing protections than an unsupported claim that stereo works on every surface.

For a proposed FAIRINO handling cell, review stereo visibility in the same layout used to select the robot. The FAIRINO buying guide helps organize the mechanical requirements; the camera trial should then verify shared visibility at the actual acquisition poses. Include the tool and container when checking for obstructions.

Request a FAIRINO vision-guided quote with working distances, bin geometry, and sample depth results. Note whether the camera is fixed or carried by the robot, and identify the untested positions the project must validate.