Machine vision / Article
Line-scan vs area-scan cameras: choosing by part speed and size
Choose an area-scan camera when discrete parts arrive one at a time and can be held still, even momentarily, inside a single field of view. Choose a line-scan camera when the material never stops, when the object is longer or wider than one frame, or when the line runs fast enough that freezing it would need impractical strobe lighting. The decision is not about image quality, it is about motion. This article works through both cases with real numbers: a continuous web, and a discrete part at takt.
The numbers at a glance
01 / The mechanism
A photograph, or a scanner
An area-scan camera holds a rectangular array of pixels and captures the scene in a single snapshot, which makes it the most common and flexible choice for inspecting discrete parts, robot guidance and general machine vision. Resolution is fixed by the sensor. Alignment is simple. If the part can be stopped or strobed, the image is complete in one exposure and nothing needs synchronising.
A line-scan camera holds one row of pixels. As the object moves past, software reconstructs a complete image line by line, which is why the useful comparison is a document scanner rather than a camera. Because there is no fixed vertical resolution, line-scan cameras can build continuous images and reach much higher resolutions than area scan, and can expose a new line while the previous one is still transferring data. The cost of that is timing: motion and acquisition have to stay locked together, or the image stretches.
02 / Worked example, web material
A roll that never stops
Take a 1,200 mm wide film web running at 500 mm per second, and a defect you need to catch at 0.1 mm. Across the width, resolution is a pixel-count question: a line-scan sensor is available from 500 to 8,000 pixels per line, with some matching acquisition rates of 67,000 lines per second, so one 8k sensor covers 1,200 mm at 0.15 mm per pixel, and two overlapping sensors comfortably beat 0.1 mm. Doing the same with area scan means tiling several cameras and stitching their seams.
Along the travel direction the arithmetic is simpler. Pixel size is object speed divided by line rate, so a conveyor at 500 millimetres per second with the camera set to 10 kilohertz gives about 0.05 millimetres per pixel. That is your figure to tune: pick a line rate that puts pixel size at or below half the smallest defect, then check the sensor can sustain it. Drive the line rate from an encoder on the roll rather than a fixed clock, so a speed change does not distort the image. This is also why web inspection favours line scan on principle: line scan handles infinite length seamlessly where area scan would require stitching.
03 / Worked example, discrete part
A connector body at 60 parts a minute
Now take a 40 mm moulded connector at a one-second takt, inspected for flash, short shots and print. It fits in one field of view, arrives in a known orientation, and only needs to be still for a few hundred microseconds. Area scan is the correct answer, and the reasoning is commercial rather than optical: for discrete parts at moderate speed, area scan offers simpler integration, lower cost and easier deployment for most discrete part inspection applications.
Geometry can still push you back the other way. If that connector needs all four sides checked, one common pattern is four area-scan cameras placed 90 degrees apart to capture a full 360 degree inspection as the product travels down a conveyor, with strobe illumination freezing the image. If the part is cylindrical and can be rotated instead, a single line-scan camera unwraps the whole surface into one flat image with no seams to reconcile. Count the cameras each option needs before you compare prices, because that is usually where the real cost difference sits.
04 / The decision
Six questions, asked in this order
One caution on cost. Buyers routinely underestimate how many area-scan cameras it takes to match a line-scan camera's resolution across a wide field of view. Do the pixel arithmetic before you compare quotations.
05 / On your line
Camera is step three, not step one
Sensor choice is where most vision conversations begin and it is the wrong place to start. At CODETRACE the sequence is lighting, then lens, then camera. Lighting comes first because a defect the illumination does not reveal cannot be recovered by any sensor: a surface scratch that disappears under diffuse light and appears under low-angle grazing light is a lighting decision, not a megapixel one. The lens comes second, because working distance, field of view and achievable resolution are fixed by optics before the sensor is chosen at all.
The camera step then has a narrow job: match what lighting and optics have made visible to the speed and geometry of your part. Getting that order right is also what keeps false calls down, because most nuisance rejects trace back to unstable illumination rather than the sensor. We run this pipeline on your actual parts before quoting, from Shah Alam in Selangor and Batu Kawan in Penang, and CODETRACE is a member of the NVIDIA Inception program. Where a defect needs measuring rather than photographing, the same feasibility work covers 2D against 3D imaging.
06 / Where to start
Bring the part, the speed and the defect
Three numbers settle most of this before anyone visits: the smallest defect you must catch in millimetres, the part or web speed in millimetres per second, and the width you have to cover. Add whether the part can be stopped and whether it can be rotated. From those five facts the sensor type is usually determined, and the remaining work is choosing lighting that makes the defect visible and proving it on real production samples rather than a clean sample from the tool room.
Do the proving before capital release. Image the parts you actually reject, at the actual line speed, with the actual surface finish, and measure both escape rate and false-call rate on the result. Inspection is one station inside factory automation, so the imaging decision should be made with handling and rejection in the same conversation. For the wider picture, our explainer on what AOI actually does across six inspection jobs shows which jobs put pressure on the camera and which put pressure on the light.
Match the sensor to the motion. Everything else follows.
FAQ / Line scan vs area scan
Questions, answered.
01What is the difference between a line-scan and an area-scan camera?
An area-scan camera has a rectangular pixel array and captures the whole scene in one exposure, like a photograph. A line-scan camera has a single row of pixels and captures one thin line at a time, stitching those lines into an image as the part or web moves past, like a document scanner. That difference decides everything downstream: area scan needs the part to hold still for an instant, line scan needs continuous, known motion.
02When should I choose a line-scan camera?
Choose line scan for continuous web materials such as paper, film, foil, textiles and nonwovens, for cylindrical surfaces such as cans, pipes and rollers where the camera effectively unwraps the surface, for large planar objects wider than one area-scan field of view, and for lines running faster than about 2 metres per second where area scan would need impractical strobe lighting. Line-scan systems deliver roughly 20 to 30 times faster inspection speeds than area scan on continuous objects.
03How do I calculate line rate for a line-scan camera?
Pixel size along the direction of travel is object speed divided by line rate. At a conveyor speed of 500 millimetres per second and a line rate of 10 kilohertz, each line covers about 0.05 millimetres, which is your resolution in the travel direction. Set the line rate so that this figure is at most half your smallest defect. In production the line rate is normally driven by an encoder on the conveyor rather than a fixed clock, so a speed change does not stretch or compress the image.
04Is area scan ever better than line scan?
Often. For discrete parts arriving one at a time at moderate speed, area scan offers simpler integration, lower cost and easier deployment, and it handles randomly oriented parts and colour accuracy well. It needs no encoder, no motion synchronisation and no stitching. If your part fits in one field of view at the resolution you need and can be strobed still for an instant, area scan is the cheaper and more robust engineering choice.
05Where does camera choice sit in a vision project?
Third. Lighting comes first, because a defect that the illumination does not reveal cannot be recovered by any sensor. Optics come second, since the lens sets working distance, field of view and achievable resolution. The camera is selected to match what the first two steps make visible, and only then against part speed and geometry. CODETRACE runs that lighting, lens then camera pipeline on the customer part before quoting, from Shah Alam in Selangor and Batu Kawan in Penang.
Related / Machine vision