Vision-Based Part Counting

Overlap significantly increases the difficulty of counting. There are two workable paths: first, improve the tooling or the way parts are tipped out so that parts overlap as little as possible; second, accept that "overlapping areas cannot be judged" and convert them into an alarm or manual review. We do not recommend promising exact counts directly in scenarios with severe overlap.

It depends on the relationship between the motion speed and the required exposure time. The higher the speed, the shorter the available exposure time and the higher the demands on illumination brightness and stability. The real usable speed limit must be measured on site, not extrapolated from a specification sheet.

Transparent parts usually need special illumination (such as backlight + polarization, or dark field) to make the edges visible; reflective parts use angle, polarization, or diffuse lighting to reduce specular reflection. The feasibility for such parts must be validated with actual imaging tests and cannot be judged from experience alone.

An extra part must also be stopped — a surplus part can cause interference, short circuits, or assembly difficulties in subsequent process steps, and must not be let through just because "at least nothing is missing". If the site only needs to know whether there are "enough" rather than an exact count, the acceptance criteria can be simplified to "no fewer than N", which lowers the demand on target segmentation and makes the implementation cost more controllable.

The difficulty in count inspection is usually not the algorithm but the separability of the targets. Parts packed tightly in the same field of view, parts overlapping each other, and parts joined into one blob by reflections all make the counting rule ambiguous. The engineering solution is usually "solve imaging first, then the algorithm": use illumination (backlight, low angle, dark field) to create a distinguishable boundary between targets, or use tooling to keep a fixed spacing between parts. Once the targets are separated, counting is a straightforward process.

View the more specific object characteristics, acceptance criteria and optical considerations by object; covers 8 common object types including screw presence/absence inspection, nut presence/absence inspection, gasket presence/absence inspection and PCB component presence/absence inspection.

What visual inspection replaces is repetitive visual judgement, not manual labor for everything. The typical division of labor is: vision performs part-by-part full inspection and judgement, while people handle re-judgement of borderline samples, changeovers and exceptions, and maintenance of the optics and tooling. How many people are actually involved depends on the degree of automation and the re-judgement strategy.

Please provide the parts to be counted and the packaging / tooling format, and explain the counting rules (for example how overlaps are handled); the engineer will assess target separability and give imaging recommendations.