Machine vision splits into two architectures. Area scan cameras - the overwhelming majority of installations - capture a two-dimensional image in one exposure. Line scan cameras capture one pixel row at a time, building the image as your product moves past. Choosing between them is less about camera specs and more about how your product physically moves through the station.
When area scan is obviously right
Discrete parts that can pause: indexed tables, robot-presented parts, stop-and-go conveyors, anything with a fixture. An area scan camera photographs the part in a single exposure - simple triggering, one image per part, algorithms that assume 2D geometry. The entire mainstream catalog, from the 5 MP MV-CS050-10GM to the 20 MP MV-CS200-10GM, lives here.
When line scan earns its keep
Continuous webs. Paper, film, foil, steel strip, textiles, printed labels - product that never stops. Stopping a web to photograph it is absurd; a line scan camera rides the motion instead, the product itself acting as the second axis of the image.
Very large or very wide fields. A web 1.5 meters wide with 0.1 mm defect tolerance needs about 15,000 pixels across. No affordable area sensor reaches that; two 8k line scan cameras do, trivially.
High-speed continuous inspection. Line scan sensors read only a row per clock, so effective rates scale with sensor design - web speeds of several meters per second with full defect resolution are routine where an area scan camera would need impractical strobing.
Unwieldy 3D geometry. Cylindrical surfaces like printed bottles or pipes are often inspected by line scan with rotary staging, capturing the unrolled surface as one flat image.
The hidden cost of line scan
Line scan systems integrate the camera into the machine kinematics. Encoder input synchronizes line rate to web speed; lighting must be a continuous line (or strobed per line) with no tolerance for variation; the lens must resolve across the full field with telecentric discipline for measurement work. Integration effort and illumination cost are both several times the area scan equivalent. And the algorithm stack differs: no single "part image", but a continuous stream to segment in motion.
A third path: the strobed middle ground
Many nominally "continuous" applications can be converted to discrete captures. Photo-eye triggers an area scan camera, a short strobe freezes motion, and the conveyor index presents each part. At 60 parts per minute with 5 ms windows, an area scan camera does the job that integrators sometimes over-spec with line scan. Run the numbers before committing: line scan is the answer when the product genuinely cannot stop, when fields are extreme, or when the web speed exceeds what strobing can freeze.
Decision summary
Part stops or can be triggered - area scan. Continuous web, extreme width, or uncompromisable speed - line scan. Large field but intermittent presentation - area scan with stitching, or larger-format sensors such as the 20 MP MV-CS200-10GM before jumping to line hardware. Every machine vision journey passes through this fork once; the ones who measure displacement, speed and field size first usually pick correctly.