This dataset tracks 65 patent families published between 2015 and mid-2026 that combine machine vision or automated optical inspection terminology with specific technical elements such as lighting design, false reject rates, line scan cameras or calibration methods. That is a filtered, purpose-built count rather than a count of every patent that mentions machine vision, so broader searches will return higher numbers. Families, not raw document counts, are the more reliable unit because they neutralise duplicate filings across jurisdictions for the same invention.

The data shows a peak of 10 families in 2019 followed by a decline to 2 by the 2022 midpoint, with no recovery through the most recent tracked years. That pattern reads as flat-to-declining rather than growing, though the last one to two years should be treated cautiously since patent publication typically lags filing by around 18 months. Anyone using this for market-sizing should pair it with product-launch or funding data rather than relying on filing counts alone.

Image data processing under IPC class G06T is the largest single category at 37 of 65 records, followed closely by material analysis and testing under G01N at 32. Optical hardware (G02B, 7 records) and dedicated image recognition (G06V, 6 records) are comparatively thin, and AI-specific classification under G06N appears in only 8 records. This suggests that claim activity in this corpus centres on processing and analysis methods more than on optical hardware or new sensor designs.

The United States leads by a wide margin with 35 of the 65 tracked records, followed by the European Patent Office at 9 and Israel at 6. China, South Korea and WIPO/PCT filings each sit in the low single digits, at 4, 3 and 3 respectively. That imbalance means an applicant checking freedom-to-operate in China or Korea may encounter less directly relevant prior art than one checking the US market.

Relative to the dense core of calibration and general defect-detection claims, sub-areas such as extended depth-of-field autofocus, line scan camera calibration workflows, deep-learning-based false-reject reduction, and inspection of transparent or reflective containers show thinner coverage. These are not unclaimed entirely, since some of the most-cited records in the corpus touch on autofocus and container inspection, but the density around them is lower than around general image-processing and calibration claims. A first claim in these areas should be scoped tightly against the existing highly-cited records rather than assumed open.