High-speed imaging for inspection

What makes imaging high-speed inspection hard

High-speed imaging is needed whenever parts move quickly past the camera, or when many discrete events must be captured per second, such as on a fast indexing line, a continuous web, a bottling or filling line, or a pick-and-place head. The core challenge is that everything competes for a fixed pool of sensor and link bandwidth: you cannot independently maximise resolution, frame rate and field of view at once.

Designing a high-speed station is therefore an exercise in budgeting. You start from the process requirement, the smallest defect to resolve, the part speed, and the throughput in parts per minute, then work backwards to a sensor, exposure, lighting and interface that all balance. Skipping that arithmetic almost always produces blurred images, dropped frames, or a system that runs at a fraction of the intended line speed.

The frame rate versus resolution trade-off

A sensor reads out a finite number of pixels per second. That pixel-throughput figure, set by the sensor and the readout electronics, is what you are spending. Doubling the pixel count of an image roughly halves the maximum frame rate at full frame, and vice versa. This is why a low-resolution sensor can run at thousands of frames per second while a high-resolution sensor may manage only tens of frames per second at full frame.

In practice you size resolution from the defect you must detect, not from a desire for sharp-looking pictures. A common rule is to span the smallest feature with roughly 3 to 5 pixels; resolving any finer wastes bandwidth you could spend on speed. Once resolution is fixed by the optics and defect size, the remaining bandwidth determines your achievable frame rate, and if it falls short you reach for a region of interest or binning before you reach for a bigger sensor.

Exposure time and motion blur

Motion blur is set by exposure time and part velocity, and it is independent of frame rate. If a part moves during the exposure, its image smears across the sensor by an amount equal to its travel divided by the system magnification. The working rule is to keep that smear below about one pixel, so blur stays under the resolution you have already paid for.

Concretely, if a feature moves at a given speed and each pixel covers a known distance on the part, the maximum exposure is the time it takes the part to travel one pixel-width. Fast lines routinely force exposures into the tens of microseconds. That short exposure is the real constraint, because it collects very little light, which pushes the whole design toward intense, well-controlled illumination rather than toward a faster camera alone.

• Estimate object-space pixel size: field of view width divided by sensor width in pixels.
• Compute travel during exposure: part speed multiplied by exposure time.
• Target blur under one pixel; if exceeded, shorten exposure and add light to compensate.
• Remember blur is a function of speed and exposure, so a higher frame rate alone never fixes it.

Strobe and overdrive lighting to freeze motion

Short exposures starve the sensor of light, so high-speed stations almost always rely on strobed illumination. Pulsing an LED in tight synchronisation with the exposure concentrates the available light into the brief window the shutter is open, dramatically increasing the effective brightness without raising average power or heat.

LED controllers built for this purpose allow overdrive, driving the LEDs well above their continuous-rated current for very short pulses. This can multiply peak output several times over, which is exactly what a tens-of-microseconds exposure needs. The trade-off is duty cycle and thermal management: pulses must be short and well spaced to protect the LEDs, and the strobe must be hardware-triggered from the camera or an encoder so the flash lands precisely within the exposure. For demanding geometry or contrast problems at speed, the lighting choice often matters more than the camera, and dedicated lighting from a brand such as DCM Sistemes paired with a proper strobe controller is the right starting point.

Why a global shutter is non-negotiable

A rolling shutter exposes the sensor row by row, so the top of the frame is captured a fraction of a millisecond before the bottom. With a stationary scene this is invisible, but a fast-moving part is recorded at slightly different positions in each row, producing skew, smear and geometric distortion that wreck measurement and defect detection.

A global shutter exposes every pixel simultaneously, capturing a single coherent instant. For any application with relative motion between camera and part, which is almost the definition of high-speed inspection, a global shutter sensor is effectively mandatory. This is one of the first specifications to lock in when shortlisting cameras.

Interface and bandwidth limits

The camera interface sets a hard ceiling on how much image data can leave the camera per second, and it is frequently the real bottleneck rather than the sensor. GigE Vision delivers roughly 1 Gbit/s over up to about 100 m with PoE, which is convenient but modest; USB3 Vision offers around 350 to 400 MB/s over short 3 to 5 m cables. When you need sustained multi-GB/s throughput, CoaXPress (CXP-6, CXP-12 and multi-link configurations) is the standard choice.

Compute your required data rate as resolution times bit depth times frame rate, then confirm the interface can carry it continuously, not just in bursts. If the link cannot keep up, the camera will drop frames or throttle its frame rate regardless of what the sensor could theoretically do. If you are weighing the options, our interface comparison guide covers the practical limits in detail.

ROI, binning and the frame grabber

When the full sensor cannot reach the needed frame rate, the two most direct levers are region of interest and binning. Reading out only the rows you need, a narrow ROI, raises frame rate roughly in proportion to the reduction in rows, which is ideal when inspecting a strip or a single feature. Binning combines neighbouring pixels, trading resolution for speed and sensitivity, useful when the short exposure also leaves you light-starved.

For sustained high throughput, a CoaXPress camera paired with a frame grabber is the workhorse architecture. The grabber handles deterministic acquisition, hardware triggering and direct memory transfer to the host, offloading the CPU and guaranteeing no dropped frames at rates a software-driven interface could not sustain. When the combined demands of speed, resolution and reliable triggering exceed what a standalone camera can deliver, that is the point to move to a grabber-based system, and where AIS Vision can help architect the acquisition chain end to end.