What goes into a vision system

Area scan cameras capture a full rectangular image (width × height) in one exposure. Every pixel in the sensor array is read simultaneously, producing a complete 2D image per trigger event. This makes them ideal for stop-and-inspect or triggered acquisition workflows.

They function exactly like a smartphone or digital camera — taking a complete snapshot in one go.

Instead of capturing a full rectangle of pixels, line scan cameras contain a single, narrow horizontal row of pixels. They build a complete 2D image row-by-row as the target object moves through the field of view — much like a document scanner stitching a copy together one strip at a time.

The capture process operates like a traditional document scanner or photocopier, building the image one narrow strip at a time.

3D cameras generate point clouds or height maps by measuring the distance from the sensor to every point on the object's surface. Different technologies achieve this in distinct ways, each with trade-offs in speed, resolution, and range.

1D systems capture and process a single row of pixel data per scan event. Each line is analyzed individually in real time — detecting defects, edges, or material variations as the web passes through at full production speed. There is no image reconstruction; the intelligence operates on the signal level.

Smart cameras combine an imaging sensor with an onboard processor running embedded vision software. They acquire, process, and analyze images internally, outputting only pass/fail decisions, measurement data, or decoded strings over industrial fieldbus protocols.

These systems use specialized sensors sensitive to specific wavelength bands — from short-wave infrared through long-wave thermal. By capturing how materials interact with non-visible radiation, they extract information about composition, temperature, and subsurface features.

Light rays enter the lens at diverging angles defined by the focal length and aperture. As an object moves further away, its magnification decreases proportionally. This perspective projection makes distant objects appear smaller relative to near objects within the same frame.

They operate exactly like the human eye or a standard camera lens. If you look down a long hallway, the walls appear to converge in the distance — that's perspective distortion, and entocentric lenses exhibit the same behavior.

By placing an aperture stop at the rear focal point, only light rays entering parallel to the optical axis reach the sensor. This parallel-ray architecture means an object's apparent size does not change as it moves closer to or further from the lens — magnification remains constant across the entire depth of field.

Imagine watching a train recede into the distance, but it never appears to shrink — it stays exactly the same size no matter how far away it gets. If you look straight down at a cylindrical pipe through a telecentric lens, you will only see the top rim, with no view of the inner or outer walls.

An electrical current applied to the lens cell changes the electrostatic pressure on the interface between two immiscible liquids (typically oil and water). This instantly alters the curvature of the liquid meniscus, shifting the focal length without any moving glass elements. Response times are typically under 10 ms.

They function almost exactly like the crystalline lens inside the human eye, which relies on ciliary muscles to instantly reshape itself and shift focus from a book in your hands to a mountain in the distance — but driven by electrical signals rather than muscle contractions.

By designing the entrance pupil to be larger than the object itself and positioning it in front of the object, light rays converge rather than diverge. This means the lens can 'see around' the object, capturing the top face and the full circumference of the side walls in one exposure without needing mirrors or additional cameras.

Imagine being able to wrap your field of vision entirely around a soup can — reading the label on all sides and seeing the top lid at the exact same time, from a single vantage point directly above.

A miniaturized lens system at the distal tip captures the image, which is relayed through the tube body via a series of relay lenses (rigid) or a coherent fiber optic bundle (flexible). Integrated LED illumination at the tip provides the necessary lighting inside dark cavities. The image is delivered to an eyepiece, camera adapter, or directly to a digital sensor.

Very similar to the medical endoscopes a doctor uses to look inside the human body — but engineered with hardened optics and housings to withstand industrial fluids, high temperatures, and harsh environments.

An annular array of LEDs surrounds the lens aperture, directing light straight down onto the object from all radial directions simultaneously. Because the light source encircles the lens, shadows cast by surface features are filled from every angle, producing extremely even illumination with minimal shadowing on flat targets.

Picture a ring of stadium floodlights arranged in a circle directly above a football pitch. Every point on the field is lit from all sides equally, eliminating the long shadows you would get from a single spotlight on one side.

A dense row of LEDs is arranged in a linear housing with integrated optics (lenses or reflectors) that collimate or focus the light into a narrow, high-intensity stripe. Multiple bars can be angled and positioned around the target to create bright field, dark field, or directional lighting geometries.

Think of the fluorescent tube lights above a workbench — except engineered to concentrate all their output into a thin, precise curtain of light rather than flooding the entire room.

A dense LED array behind a diffusion panel produces an extremely uniform, flat illumination surface. The object is placed between this light source and the camera, blocking the light and creating a crisp, high-contrast shadow. The camera sees only the object's outline against a bright, featureless background.

The same principle as shadow puppets — you place an object between a bright wall and your eyes, and you see a perfectly sharp outline. The brighter and more uniform the wall, the sharper and more precise the silhouette.

LEDs mounted inside a hemispherical housing project light upward onto the dome's white interior surface, which acts as a diffuse reflector. The light bounces down onto the object from every conceivable angle simultaneously. A port at the top allows the camera to image through the dome. This geometry is sometimes called "cloudy day" illumination.

Imagine placing an object under a perfectly overcast sky — light arrives equally from every direction with no direct sun, so there are no harsh shadows or glare spots. A dome light recreates this effect in a compact, controlled enclosure.

The light source is positioned at a very low angle relative to the surface plane (typically < 15°). On a smooth, flat surface, the light reflects away from the camera, leaving the image dark. But where there is a scratch, raised edge, contaminant, or surface texture change, the light is scattered upward toward the camera, making the defect appear bright against the dark background.

The same effect as running a flashlight across a wooden table at a very low angle late at night. The surface looks dark and uniform, but every scratch, dust particle, and grain imperfection suddenly glows brightly because it scatters the grazing light toward your eyes.

An LED light source projects through a half-mirror (beam splitter) mounted at 45° between the lens and the object. The light travels straight down along the optical axis, reflects off the flat, specular surface directly back through the beam splitter into the camera. Any surface deviation — a scratch, etched mark, or height change — scatters light away from the axis and appears dark against the bright, uniform background.

Like looking straight down at a calm lake — you see a perfect, bright reflection of the sky. But any ripple, oil slick, or floating debris disrupts the mirror surface and becomes immediately visible as a dark disturbance.

A laser or LED projector casts a known geometric pattern onto the target. An angled camera observes how the pattern deforms across the surface. Triangulation algorithms convert the measured distortion into a dense 3D point cloud or height map. For laser line profiling, a single line sweeps across the object as it moves, building a 3D profile scan line by line.

Imagine projecting a perfectly straight grid of lines from an overhead projector onto a crumpled piece of paper. Where the paper rises or dips, the lines bend and distort. By measuring exactly how each line bends, you can reconstruct the 3D shape of the paper's surface.

Different materials absorb and reflect light differently at each wavelength. By selecting a wavelength that maximizes the contrast between the feature of interest and the background, defects or markings that are invisible under white light become clearly visible. UV lighting additionally excites fluorescent materials, making coatings, adhesives, and contaminants glow brightly.

Like using a blacklight at a crime scene — substances invisible under normal room lighting suddenly fluoresce and glow vividly when exposed to ultraviolet light, revealing exactly where they are.

An engineer manually defines strict processing steps: find edges using gradient operators, measure the pixel distance between Edge A and Edge B, compare the result to a tolerance band. Every decision is an explicit mathematical operation with no ambiguity. The system either passes or fails a part based on exact numerical thresholds.

Like using a rigid metal stencil to check a manufactured part. If the part fits perfectly inside the stencil, it passes. If the part is slightly warped or the lighting shifts a shadow, it fails immediately because it doesn't match the exact mathematical rules.

A neural network is fed labeled training images ("good" and "bad" examples). Through iterative optimization, the network adjusts millions of internal parameters to learn statistical patterns that correlate with each class. Once trained, it classifies new, never-before-seen images by recognizing the same learned patterns — generalizing from examples rather than following explicit instructions.

Like training a new employee on an assembly line by handing them a stack of flashcards. You show them 50 pictures of bruised apples and 50 pictures of healthy apples. Eventually they develop an intuitive "gut feeling" for what a bruise looks like, even if the next bruise is a completely different shape.

A series of convolutional layers apply small, learned filter kernels (typically 3×3 or 5×5 pixels) that scan across the entire image. Early layers detect primitive features like edges and gradients. Deeper layers combine these into increasingly abstract representations — corners become shapes, shapes become objects. Pooling layers reduce spatial resolution between convolutions, making the network efficient and translation-invariant.

Imagine inspecting a massive wall painting by holding a small magnifying glass and scanning it inch by inch, top to bottom. You note down local details — a brush stroke here, a color gradient there — and piece them together later to figure out what the entire painting depicts.

The image is split into a grid of fixed-size patches (e.g., 16×16 pixels). Each patch is flattened into a vector and passed through a transformer encoder. The self-attention mechanism computes a weighted relationship between every pair of patches, allowing the network to understand how a feature in one corner of the image relates to a feature in the opposite corner — in a single layer, without sequential scanning.

Imagine cutting a photograph into a jigsaw puzzle and spreading all the pieces on a table. Instead of examining one piece at a time, you look at every piece simultaneously and instantly understand how a dark piece in the bottom-left corner relates to a bright piece in the top-right corner.

When an edge falls between physical pixels, the camera records a gradual brightness transition across several pixels rather than a sharp step. Sub-pixel algorithms model this blur profile mathematically — fitting a curve (typically Gaussian or error function) to the intensity gradient — and calculate the precise fractional pixel position where the true edge lies.

Imagine trying to measure the exact edge of a fuzzy, blurry shadow using a school ruler. Instead of just picking the grayest part, you use an advanced math formula to calculate exactly where the light ends and the dark begins — giving you a measurement far more precise than the lines on your ruler.

A reference template (or "model") of the target is created from a training image. The algorithm encodes the geometric features of this template — edge gradients, contour shapes, spatial relationships — into a compact descriptor. At runtime, it searches the full image for regions that best match this descriptor, scoring candidates by geometric similarity rather than raw pixel intensity.

The machine vision equivalent of playing "Where's Waldo?" — you know exactly what Waldo looks like, and you scan a chaotic, crowded scene to find him, even if he's standing upside down or partially hidden behind a building.

The software provides a library of pre-built processing blocks, each representing a specific algorithm or I/O operation. Users drag blocks onto a visual canvas and connect them in sequence — defining the data flow from image acquisition through processing to pass/fail output. Parameters are configured through graphical dialogs rather than code. The underlying engine compiles and optimizes the flowchart for real-time execution.

Like building a complex machine out of digital LEGO blocks, or drawing a flowchart on a whiteboard. You snap the processing steps together in the order you want them to happen — no programming degree required.