Zivid vs Photoneo vs Mech-Mind: which 3D camera for bin picking?

Why sequential patterns need a still scene

Most structured-light cameras, including Zivid's and Photoneo's PhoXi scanners, project a short series of patterns and combine them. That's how they get dense, accurate clouds. The cost is time: if the part moves during the sequence, the patterns don't line up and the cloud smears. Photoneo lists PhoXi scanning times starting at 250 ms, and Zivid quotes typical captures of 25 to 1,500 ms including processing, depending on settings. For a part sitting in a bin, that's fine. For a conveyor, it isn't.

What Photoneo's parallel structured light changes

Photoneo describes MotionCam-3D's parallel structured light as patented, and says it scans in real time without stopping the line, at up to 20 fps and for objects moving up to 40 m/s. The specs list a 10 ms data acquisition time on the S, M and M+ models. Photoneo doesn't publish how the sensor does it in detail, and we won't guess. What matters in practice is that you can scan a moving tote or keep the robot moving through the capture.

There's a trade. MotionCam-3D's depth map tops out at 1680 × 1200, about 2 MP, while the Zivid 2+ models run a 5 MP sensor according to distributor STEMMER IMAGING. For a still bin of small parts, more points usually wins. For a moving scene, a slightly coarser cloud that isn't smeared wins every time.

Read the spec words carefully

This is where most comparisons go wrong. Zivid quotes "point precision" and "trueness." Photoneo quotes "calibration accuracy" and "temporal noise." Mech-Mind quotes "Z repeatability" on a ceramic plate and accuracy to VDI/VDE 2634 Part 2. Some of those describe noise, some describe bias, and they're measured at different distances. Putting Zivid's 80 µm next to Mech-Mind's 0.1 mm and calling one "better" is comparing a thermometer's jitter with its calibration error.

For bin picking, both matter, but differently. Noise decides whether the matching algorithm can find a part's edges at all. Bias decides whether the gripper lands where the cloud says the part is. The bake-off at the end of this page measures both on your parts, which is the only number we'd actually trust.

Zivid: camera-first, now with motion planning

The Zivid SDK supports C++, .NET and Python, plus ROS (with ROS 2 Jazzy, Iron and Humble drivers) and experimental GenICam and HALCON support, on Windows, Ubuntu and NVIDIA Jetson Orin. It includes in-field camera correction and 3D hand-eye calibration. In March 2026 Zivid added Zivid Motion, a deterministic motion planner it says returns paths in under 100 ms and works with any robot brand. That's a real step toward a fuller stack, but detection and grasp logic are still mostly yours, or a software partner's.

Photoneo: a picking application in the box

Bin Picking Studio is Photoneo's ready-made picking application. Photoneo says you can set it up in 20 minutes and lists ABB, KUKA, Yaskawa, FANUC, Stäubli, Kawasaki, Universal Robots, Mitsubishi, Nachi, Denso, Doosan and Robco on its robot page. AnyPick adds AI picking of mixed items from 1 × 1 cm without CAD models, which Photoneo rates at up to 500 picks per hour. Both can be bought or leased.

Mech-Mind: the whole stack, code-free

Mech-Vision is code-free vision software with a library of more than 1,000 robot models; Mech-Mind says a machine-tending project can be set up in under 10 minutes. Mech-Viz plans and simulates collision-checked robot paths from a flowchart. Mech-DLK handles deep learning and Mech-MSR handles 2D and 3D measurement. If you'd rather code, the Mech-Eye SDK has interfaces for C++, C#, Python, LabVIEW, VisionPro, HALCON, ROS and ROS 2.

In practice "supports FANUC" can mean anything from a socket example to a fully tested interface with a pendant plugin. So ask each vendor to show the actual integration running on your robot brand and controller generation, not a slide with logos. If you're running UR, FANUC CRX or another cobot, ask specifically about payload, since a 3D camera mounted on the wrist eats into it. The Zivid 2+ weighs 1,000 g, Photoneo's MotionCam-3D M+ 1,050 g, and Mech-Mind's NANO ULTRA 0.7 kg.

Hand-eye calibration is the other thing to see in person. Every one of these cameras can be calibrated to the robot, and every one of them drifts out if the bracket gets bumped. Ask how long recalibration takes and whether a technician can do it without the vendor on the phone. For the robot side of the cell, our vision-guided robotics guide walks through calibration and the common failure modes.

We only publish prices we can source, and none of these vendors posts one. A camera that looks cheaper can come with less software, so you buy a picking package separately or pay an integrator to write it. A camera that looks expensive may include detection, grasp planning and a robot interface you'd otherwise build.

So quote the same scope from all three: camera, software licenses for vision and path planning, calibration tools, mounting, and the engineering hours to reach your pick rate on your parts. Then compare. Our breakdown of machine vision system cost shows the usual split, and how much an industrial robot costs covers the rest of the cell.