Cameras / depth vision

Images and depth maps locate objects, guide manipulation and reveal obstacles. Placement, calibration and lighting shape the usable view.

Inside a stereo depth camera

Two infrared cameras estimate depth. A projector adds texture for matching, while the RGB camera records colour.

Drag to rotate · scroll to zoom

Loading CAD model…

RealSense D435

Select a part to inspect its role.

Technical reference
Source & model scope

Official RealSense ROS geometry, with the optical apertures labelled in their original positions. Colour identifies each optical function. Optics view dims the enclosure and labels the front apertures. Rotate, isolate or focus a part to inspect it; electronics remain exterior geometry.

Original design · Apache 2.0 licence · Conversion notes · Upstream notices

Representative products

Component directory
Architecture, materials & specificationsArchitecture options, material requirements and integration constraints

Architecture options

ApproachMechanismAdvantagesTrade-offs
Passive stereoMatches corresponding features in two calibrated images.Estimates depth without a projector.Plain surfaces may have too little texture; darkness and reflections also cause problems.
Active stereoAdds projected texture to help the two cameras match features.Improves matching on surfaces with little natural texture.Projector power, working range and ambient illumination limit the benefit.
Time of flightEstimates distance from the travel of emitted light.Does not need image features for stereo matching.Indirect reflections, target reflectivity and ambient light can distort depth.

Specifications & operating limits

Image resolution counts pixels; depth accuracy describes distance error. Field of view is the visible angle, and minimum working distance is the nearest useful range. Frame rate counts captures, whereas latency includes the time needed to deliver and process them.

Materials & constraints

Optical glass & windows
Transmit and focus light at the required wavelengths.
Image sensors
Convert light into electrical measurements.
Boards & housing
Carry processing, connections and mechanical protection.

Dependencies & engineering limits

Assembly & calibration
Stereo depth depends on the relative positions and optics of two cameras. Camera calibration measures that geometry, and depth checks quantify the assembled system’s error.
Mounting & depth quality
Mechanical shifts can disturb camera geometry. Mount protection and calibration checks after impact or repair preserve the camera geometry.
Structure & materials
Scene & missing measurements
Plain, shiny or transparent surfaces can defeat depth estimation. Missing or unreliable depth is different from a measured surface; that distinction affects how the robot interprets obstacles.
Images & bandwidth & delay
Higher resolution and frame rate create more data to move and process. Synchronisation and processing latency determine how current the robot’s view is when it acts.
Compute & control

Manufacturing & assembly

Guide

About this route

The route shows one process or architecture. Select a step to see its function and connections; related parts can be parallel inputs or alternatives.

Function

Inputs & dependencies
Outputs & connections
Related processes & parts
Sources

Integration / calibration · Representative route. Finished lenses and sensors join before calibration of the pair.

Related parts & processes 5

Lenses & image sensors

A stereo pair captures two views of the same scene.

Both cameras’ optics and positions matter.

Sources & technical scope

Illustrative images of generic materials and components.

Related references
Link to this detail

Production equipment

Market context

Upstream image sensors

≈50%Sony · global CMOS sensor revenue

Sensor chips are a different market from complete depth cameras or robotics perception.

Robotic perception market

US$600m2026 market estimate
US$1.6bn2030 forecast

Cognex’s estimate spans robotics applications; it is not a humanoid-camera market size.

RealSense

80%Humanoid adoption · company claim
60%AMR adoption · company claim

Company claim; the measurement period, denominator and counting method are not disclosed.

2 more updates

Cognex agreed to acquire RealSense for approximately US$500 million in cash, extending its industrial machine-vision business into robotic depth perception.

RealSense plans to begin D585 Pro depth-camera shipments in Q1 2027. The camera integrates a new onboard processor for depth and image processing.

RoboSense

RoboSense scheduled initial Space Camera deliveries by the end of Q3 2026 and full-scale production in 2027.

Orbbec

Orbbec broke ground on a Vietnam factory for 3D sensors and intelligent hardware, targeting production in 2027. Its Foshan factory is already ramping. H1 2026 sales of vision systems to robot customers more than doubled year on year.

LG Innotek

TDK and LG Innotek agreed to develop vision and tactile modules for humanoids, combining optical-module design with inertial, position and other sensors.

Ouster

Ouster reports acquiring Stereolabs on 4 February 2026, bringing ZED cameras, digital LiDAR, embedded compute and perception software into one supplier group.

Explore documented supply relationships →