About
NOCTVISION is a technical record site covering machine vision and edge AI.
What This Site Covers
This site organizes the problems encountered when building vision systems in industrial settings — the points where optics, lighting, and hardware are entangled in ways that algorithms alone can’t resolve.
- Optics & Lenses — lens selection, focal length and depth of field, F-number, aberrations
- Lighting Design — choosing a lighting method, wavelength and reflection control
- Vision Algorithms — preprocessing, inspection techniques, applying deep-learning models
- Edge AI — model lightweighting and on-device inference optimization
Three Principles
This site tries to avoid abstract conceptual writing or unsupported claims that something “works well.” The following three principles are the standard.
Principle 1 — Optics Is Physics
The first step in system design is the lens and the lighting. This site covers physical quantities and formulas that can be verified — focal length, aperture value, depth of field. Units are unified in millimeters.
Principle 2 — Only Verified Material
Code that hasn’t been confirmed to work isn’t posted. Code includes comments on the core logic and the related physical formulas so a reader can reproduce it.
Principle 3 — Edge Environments Are the Baseline
The standard is what actually runs on an edge board, not the ample resources of a PC. Results are presented as numbers — processing speed, inspection resolution, accuracy.
Who This Is For
- Vision inspection developers — how to choose a lens by calculation rather than intuition; how focal length and F-number are derived from field of view, working distance, and sensor size.
- Those starting out with edge AI — working through, in numbers, the situation where something “worked on the PC but not on the board.”
- Engineers who want to properly understand optics — MTF, the diffraction limit, and the reason telecentric lenses exist, treated as calculable physics rather than memorized facts.
Why Keep This Record
Machine vision is easily misunderstood as simply mounting a camera and adding an algorithm. In practice, a single lighting angle or a single lens choice can overturn the result. That kind of content rarely appears in equipment catalogs or papers.
Organizing what’s been confirmed here should save time for anyone facing the same problem.
About the Articles
The articles here are written from general technical principles and publicly available material. Specific companies’ equipment specifications, process conditions, or client information are not covered.
The formulas and figures posted are for reference purposes. Please validate independently before applying them to an actual system design.
Contact
If you have a correction to point out or a topic you’d like covered, please let us know through the Contact page.