Short answer: structured light projects a pattern and reads its distortion, giving the best short-range precision on cooperative surfaces; stereo vision triangulates from two cameras like human eyes, giving cheap, sunlight-tolerant depth that depends on scene texture; time-of-flight (ToF) times returning light pulses, giving fast, compact depth that struggles with reflective multipath. Choose by how each technology fails in your scene, not by headline resolution.
Structured light: precision with conditions
A structured-light camera projects a known pattern — stripes, dots or a coded grid — and computes depth from how the pattern deforms across the object's surface. Because the camera supplies its own texture, it works on blank, featureless parts that defeat stereo, and at close range it delivers the finest point clouds of the three: this is why it dominates bin picking and metrology-style inspection. The conditions are real, though. Projection takes time — multi-shot patterns blur on moving parts — ambient light competes with the projector, and dark or mirror-like surfaces return poor signal. Our guide to 2D vs 3D vision covers when depth sensing pays at all; structured light is the answer when the scene is controlled and the tolerance is tight.
Stereo vision: the texture bargain
Stereo cameras place two imagers a known distance apart and triangulate matching points between the views. No projector, no moving parts, works outdoors and at longer ranges — and the depth is only as good as the matching. Textureless walls and repetitive patterns starve the matcher, which is why many stereo products add a pattern projector for close work, quietly becoming structured-light hybrids. The Intel RealSense line defined this category for a generation of developers and robot prototypes; Intel spun RealSense out as an independent company in 2025, a sign of how central depth cameras have become to robotics roadmaps.
Time-of-flight: speed with ghosts
ToF sensors emit modulated or pulsed light and measure the return time per pixel, producing depth at high frame rates from a compact, solid-state package. That makes ToF the natural choice for navigation, obstacle avoidance and people detection, where a fast, good-enough depth map beats a slow perfect one. Its weakness is multipath: light bouncing off a shiny floor or a metal corner arrives late and reports phantom depth, and range falls quickly against dark surfaces. Industrial suppliers in the directory's machine vision and 3D sensing category (54 companies as of 10 November 2026), among them Basler with its industrial ToF cameras, publish multipath and ambient-light behaviour precisely because experienced buyers ask.
| Structured light | Stereo | Time-of-flight | |
|---|---|---|---|
| Principle | Projected pattern deformation | Two-view triangulation | Light return timing |
| Needs scene texture | No — supplies its own | Yes, or an added projector | No |
| Best range | Short (benchtop to a few metres) | Medium to long | Short to medium |
| Moving scenes | Weak if multi-shot | Good | Good |
| Typical failure | Ambient light, dark/glossy surfaces | Textureless or repetitive scenes | Multipath reflections, phantom depth |
| Typical robot jobs | Bin picking, measurement, inspection | Navigation, research platforms | Obstacle avoidance, pallet and people detection |
Hybrids, and a word about lidar
Product labels increasingly hide hybrids. Stereo cameras ship with pattern projectors for close-range texture; ToF sensors pair with high-resolution colour cameras whose images sharpen depth edges; structured-light systems add a second mode for larger scenes. This is healthy engineering, but it means the datasheet's technology label tells you less than it used to — ask which principle produces the depth in the mode you will actually run. Lidar sits adjacent to all three: it also times returning light, but scans point by point over long ranges for navigation rather than imaging a work zone, and it is best evaluated as its own category.
Matching technology to task
Start from the scene, not the sensor. Controlled lighting, static parts, tight tolerance: structured light. Outdoor or long-range navigation: stereo (or lidar, its own category). Fast depth around people and forklifts: ToF, with its multipath behaviour tested against your actual floors and racking. In every case, the vendor demo to request is your worst part in your worst lighting — the technologies differ most exactly where demos are usually staged least.
Frequently asked questions
Which 3D camera technology is most accurate?
At short range in controlled conditions, structured light generally delivers the finest depth data, which is why it dominates bin picking and measurement. Accuracy is scene-dependent, though: ambient light and glossy or dark surfaces degrade it, while stereo and time-of-flight have their own error modes. The right question is which technology is most accurate on your parts, in your lighting, at your working distance.
Why does stereo vision fail on plain surfaces?
Stereo computes depth by matching the same point in two camera views. A blank, textureless surface gives the matcher nothing distinctive to lock onto, so depth becomes noisy or drops out. Many stereo products therefore include a pattern projector to add artificial texture at close range — effectively borrowing the structured-light trick.
What is the multipath problem in time-of-flight cameras?
A ToF pixel assumes its light travelled straight out and back. On reflective floors, metal corners and glossy packaging, light can bounce off another surface first and arrive late, so the camera reports depth that is not there — phantom points in the scene. Testing a ToF camera against your real floors and racking matters more than its laboratory range figure.
Can one 3D camera serve both picking and navigation?
Sometimes, but it is a compromise. Picking wants short-range precision (structured light territory), while navigation wants fast, longer-range awareness (stereo or ToF). Many production cells separate the two: a high-precision camera over the pick point and cheaper depth sensing for the robot's surroundings. Decide from the failure cost in each role, not from the parts list.