How we made Zivid Two - our 2nd generation 3D machine vision camera for robotics
summary
Zivid Two (CV2) is a robot‑mounted 3D camera built from fully custom optics, electronics and projector to halve size/weight, deliver ~3× speed and accuracy versus the prior product, and operate at 100 ms capture while surviving shock, vibration and robot accelerations. Design uses an FPGA for on‑sensor throughput and an SDK (C++11 with Python/.NET wrappers). Thermal and mechanical drift stays under 0.1 pixel (~500 nm) after calibration; launch slips from Nov 2020 to shipping in May.
AI-written summary · about
chapters (22)
- 0:00Zivid and CV2 intro
- 2:36Robotics growth reasons
- 4:24Human-like robot vision
- 7:23Fixed-position camera problems
- 10:13Robot-mounted camera benefits
- 13:02CV2 design goals
- 14:46Custom components for performance
- 16:29Structured light technology
- 16:53Thermal and mechanical stability
- 21:51Industrial design features
- 22:56Hardware engineering
- 25:08Iterative testing
- 26:54Optics and projector innovations
- 27:17Software stack and SDK
- 29:48Cross-platform API tools
- 31:58Modern software practices
- 32:19Comprehensive testing
- 35:51Mechanical stress testing
- 38:03Performance vs competitors
- 39:31CV2 real-world uses
- 41:41Project retrospective
- 44:29Culture and innovation
related talks
23mComputer Vision Toolbox - Connected Computer Vision & AI Quality Inspection at Scale!
41mComputer Vision for Remote AR
27mAutomated Python Powered Turret
46mChallenges of processing high-volume camera data streams
28mVision, Big Data and AI
10mGenesis Final Round - STUDIO LAB
20mImplementing Vision AI in Enterprise Android Apps
25mLibcamera: Making Complex Cameras Easy