Key Takeaways
- 2D vision relies on contrast and lighting, which can limit reliability when conditions vary or geometry drives inspection.
- 3D vision measures height, depth, and surface profile to detect defects across color, sheen, and orientation changes.
- Manufacturers use 3D vision to reduce false rejects, redeploy labor, and recover yield from harder inspections.
You aren't really inspecting your cylindrical parts; you're hoping your software can guess what's hidden in the blur.
That might sound harsh, but if your production line still relies exclusively on 2D machine vision for complex geometry or variable-component inspection, you're making quality decisions on a flat, intensity-based shadow of reality.
This post breaks down where 2D vision hits its ceiling, why 3D was built to break through it, and how operations teams use 3D vision to recover yield and redeploy manual inspection labor.
What Is 2D Vision, and Where Does It Break Down?
A 2D vision system inspects parts by capturing a flat image and analyzing pixel-by-pixel variations in contrast, brightness, and pattern. When conditions are consistent and the part is simple, 2D performs reliably. The problem starts the moment your environment or material properties introduce variability.
Why Does Lighting Determine Everything in a 2D Vision System?
In 2D machine vision, lighting doubles as the inspection itself. Because a 2D system reads contrast, measurement accuracy depends entirely on how light falls on the part, and even a perfect part can trigger errors from lighting variation alone.
That constraint compounds on real lines: color changes between runs, finishes vary across batches, and ambient light shifts between shifts, so no configuration handles every case. As David Wyatt, founder of Automation Doctor, told Vision Systems Design:
“Because the product changes so much, you cannot find a stable lighting scenario that works for more than a couple of days.”
When you can't stabilize lighting, 2D vision can't stabilize its results.
What Can't a 2D Vision System Measure?
2D sensors don't support shape-related measurements: flatness, surface angles, and part volumes all fall outside what a flat image can capture. You can't tell if a solder dot is properly domed from its top, or confirm an adhesive bead is volumetrically complete from its footprint.
But there's a subtler failure mode that's easy to miss: 2D can measure the right thing accurately and still miss the defect. JingGuo Zhang, Principal Applications Engineer, Customer Success at Cognex, describes this from a recent engagement:
A customer was inspecting whether a cable was correctly plugged into a connector slot in an EV control box. The connector was horizontal, so they used a 2D smart camera looking down and measured the distance from the cable terminal block feature to the slot feature. But when the cable was plugged in at an angle, up or down, the distance still showed within tolerance, though it wasn't tightly connected. They spent two weeks trying to fix it, tilting the camera, changing the lighting, even considering a second side-looking camera, but installation space limited every option. We recommended they use a 3D camera to measure the surface plane angle of the cable terminal block and determine whether the cable was connected level. That solved it immediately, because it directly measured the cable's 3D position.
That delay was the real cost: the 2D system wasn't broken. It measured exactly what it was designed to measure, and it still produced a wrong answer, because the defect lived in a dimension the camera couldn't see.
3D vision can inspect flatness and warping where 2D vision systems can’t.
3D vision excel at inspecting parts with complex curved geometries.
The following table shows where 2D and 3D vision are a good fit for various inspection tasks.
Inspection Task | 2D Vision | 3D Vision |
|---|---|---|
| Presence/absence detection | ✅ | ✅ |
| Surface color or print defects | ✅ | ✅ |
| Height measurement | ❌ | ✅ |
| Flatness / warp detection | ❌ | ✅ |
| Volumetric inspection (e.g. adhesive, solder) | ❌ | ✅ |
| Inspection of reflective surfaces | ❌ | ✅ |
| Part inspection without fixed lighting | ❌ | ✅ |
| Complex geometry (e.g. cylinders, curves) | ❌ | ✅ |
What Is 3D Vision, and How Does It Actually Work?
3D machine vision captures a part's true geometry, including height, depth, and surface profile, rather than just a flat image. Instead of inferring shape from contrast, 3D measures it directly, producing a point cloud or height map that quantifies every spatial dimension.
Several core technologies drive 3D industrial vision:
- Laser triangulation: a laser line projects across a moving part, building a 3D profile line by line, ideal for inline inspection on a conveyor.
- Structured light: a known light pattern reveals surface geometry through its distortions, delivering high-resolution models that excel at spotting scratches and dents.
- Stereo vision: two offset cameras calculate depth from image disparity, working well for bin-picking and sizing in less controlled lighting.
- Time-of-flight (ToF): compact and fast, ToF measures how long projected light takes to return, though with lower precision than structured light.
How Does 3D Vision Solve the Lighting Problem?
Because 3D systems build geometry from physical measurement rather than pixel intensity, they aren't tied to a fixed lighting setup. The output is a spatial data set, not a contrast map, so 3D stays robust as color, finish, or ambient light varies.
That said, environmental lighting still matters, a detail Zhang says most practitioners overlook. Zhang sees it this way:
Environmental lighting can affect 3D camera imaging, especially for stereo 3D cameras. Glare on the part can overwhelm the projected 3D lighting feature, preventing point cloud generation.
For reflective surfaces, advanced 3D solutions use high-powered, speckle-free lasers that maximize contrast and suppress glare, enabling accurate inspection on shiny metal or polished parts where 2D would blow out highlights.
Where Does 3D Vision Actually Win on the Production Floor?
3D machine vision solves problems 2D fundamentally can't handle, most often volumetric inspection, complex geometry, and variable surface conditions.
Adhesive and Sealant Inspection
RTV silicone sealant protects circuit boards and connectors from vibration and temperature stress. It must go on within tight tolerances: too much sealant risks a short circuit, too little leaves components exposed.
A 2D camera looking down at a sealant bead sees only its footprint, so it can't tell if the bead is properly domed or contains voids. A 3D system measures height across every point, quantifying shape and volume against tolerance.
Low-contrast Surface Inspections
How Does BorgWarner Use 3D Vision to Inspect EV Components
BorgWarner deployed Cognex 3D machine vision at its Changyeong, South Korea facility to inspect EV power drive assemblies, after low contrast made reliable 2D detection of missing parts impossible, even with repositioned cameras or adjusted lighting.
2D systems, per Assembly Magazine's August 2025 coverage, “need highly controlled environments with standard viewpoints and lighting that creates high contrast and eliminates shadows.” 3D vision resolved what 2D couldn't, and the same deployment confirmed sealant integrity, catching voids 2D missed. Processes that once needed five operators now run on one Cognex system, freeing that labor for higher-value work.
Bin Picking and Robotic Guidance
Robotic bin picking demands vision-guided automation because parts arrive randomized in position and overlap. A 2D camera can confirm a part is present but picking a specific one from a cluttered bin needs depth data and pose estimation that 2D can't supply.
BOS Innovations, a Canadian automation integrator, first tried a bin-picking application for a nuclear components manufacturer with 2D vision and a laser:
“We tried solving the bin-picking application with 2D machine vision and a laser, but we did not see the level of robustness we would be proud of,” said Alex Klarenbeek, Senior Project Lead for Vision Systems at BOS Innovations.
The team switched to a Cognex 3D line scan camera, which produced high-resolution imagery on specular surfaces and gave the robot a precise pick target.
Does Your Application Actually Need 3D?
3D machine vision is the right answer for specific problems, not a universal replacement for 2D, and the industry sometimes over-corrects, particularly with narrow but deep defects, as Zhang explains:
When customers switch from 2D to 3D to detect very thin but deep defects, they may encounter the issue that 3D cannot detect the defect well, especially when the surface is not very flat. This is because 3D imaging has limitations when imaging thin defects: laser or structured LED light cannot be projected into the inner part of a thin defect. In such a case, 2D combined with 3D is a good option.
3D vision reads surface geometry, not sub-surface structure, so narrow cracks deep in a part may not distort the surface enough to resolve reliably. Combining 2D and 3D, letting each do what it does best, often beats either alone.
3D is the clear choice when:
- You can't stabilize your lighting: variability that challenge a reliable 2D setup is no longer a constraint.
- You need volumetric data: flatness, warp, and bead volume all live in the third dimension, where 2D can't find them.
- Your parts are reflective: shiny metal and polished glass make contrast-based imaging difficult, exactly what laser triangulation is built for.
- You're guiding robots to unpredictable positions: bin picking and flexible assembly need spatial data 2D can't supply.
2D vision remains the right call for high-speed surface inspection, label verification, and stable, controllable parts.
How AI Changes the 3D Vision Equation
Modern 3D systems don't run on rigid rule-based toolsets alone: embedded AI learns what “acceptable” looks like from real production samples, then flags anomalies outside those norms.
Cognex 3D vision systems pair 3D imaging with AI-powered machine vision software that detects variable or undefined features, transforming 3D data into projection images optimized for AI-based training. That cuts setup time dramatically versus rule-based programming, and removes the barrier that kept 3D vision out of smaller, mid-complexity operations.
What Should Manufacturers Do Differently Starting Now?
The most actionable shift isn't a technology purchase; it's a change in process sequence. Zhang puts it simply:
With 3D widely used, I hope customers will consider 3D vision at the very first feasibility investigation period. 3D is the most direct representation of our 3D world. 2D is just a projection of 3D, and information will be lost when acquiring a 2D image.
Starting with the technology that captures the world as it exists, in three dimensions, changes the economics of an inspection project. It eliminates the troubleshooting cycles and compromised thresholds that come from forcing 2D to do a 3D job.
When inspection results are inconsistent, the issue often lies in trying to solve a dimensional problem with flat-image data. The right vision approach starts by matching the inspection challenge to the measurements the system can capture.
Ready to see the full picture?
Download the Cognex 3D Machine Vision Applications Guide to see how 3D vision can deliver high-accuracy inspections, measurements, and robotic guidance for manufacturing.