In industrial applications such as new-energy battery testing, smartphone screen defect detection, and automotive component measurement, machine vision systems serve as the cornerstone of quality control. And while image acquisition cards—often referred to as the system’s “unsung heroes”—don’t directly “capture images” like cameras or “make judgments” like algorithms, they do play a crucial role in converting camera signals into data that computers can process, making them nothing less than a “data bridge.”

Many people mistakenly believe that “cameras can directly transmit images to a computer,” but this is actually not the case: Without a capture card, analog camera signals will appear as “garbled text”; high-speed digital cameras will experience “data congestion”; and when multiple cameras are used to synchronize their footage, the images will become “out of sync.” In this article, we’ll break down the core logic of this technology from four perspectives—definition, principle, parameters, and applications—to help you understand why it’s an essential component in machine vision.

I. A New Perspective: The image acquisition card is a “translator + controller.”

When people hear the term “capture card,” they’re likely to think of “storage cards.” However, in the field of machine vision, an image capture card serves as a “two-way interactive hub” that connects industrial cameras to computers. Its core functions fall into two main categories:

1. Signal “Translation”: Enabling seamless communication between devices

Industrial camera output signals fall into two categories, both of which require a capture card to convert the format:

· Analog signal (traditional CCD camera)It outputs a continuous electrical signal, similar to “analog broadcasting.” However, computers can only process digital signals. The acquisition card uses “analog-to-digital conversion (A/D conversion)” to transform the 0–5V signal into digital values—such as 0–255 or 0–4095—with an accuracy of 8-bit or 12-bit, enabling the computer to “understand” the image.

· Digital signal (mainstream industrial cameras)Output in vendor-specific formats (such as the Camera Link protocol), which cannot be directly parsed by the operating system. The acquisition card acts as a “decoder,” converting the data into universal formats like BMP or RAW before transmitting it.

Case: A bottle cap manufacturer uses a simulated CCD camera to detect scratches. When connected directly to a computer, the camera only displays voltage fluctuations. After inserting a data acquisition card, the signal is converted into a 12-bit digital image, making 0.1mm scratches clearly visible. The computer can then precisely mark defective products.

2. Camera “Control”: Enhancing Image Capture Accuracy

In addition to signal conversion, the acquisition card can also “control” the camera:

· Trigger controlIn the production line, after the receiving sensor detects the “part in place” signal, it triggers the camera to expose, thereby preventing “empty shots” and “missed shots.” For example, on a food packaging line, the camera takes a picture only when the packaging bag is properly positioned.

· Parameter adjustmentUsing software, remotely set the exposure time, gain, and other parameters. For detecting reflective metal parts, shorten the exposure time to prevent overexposure; for capturing transparent plastic parts, increase the gain and enhance contrast.

· Multi-device synchronizationDuring 3D detection and panoramic stitching, a synchronization signal is sent to ensure that multiple cameras “expose simultaneously.” The synchronization error can be controlled within 0.1 microseconds, thereby preventing deviations in 3D modeling.

II. Decomposition Principle: Complete “Signal Reception - Processing - Transmission” in 3 Steps

The image acquisition card workflow can be broken down into 3 steps, designed around “efficiency and precision”:

1. Signal Reception: Get Ready for the Handover

After the camera is powered on, the capture card performs a “protocol handshake” with the camera to confirm parameters such as model, resolution, pixel clock, and initialize the buffer. If connecting to a GigE camera, verify the frame rate and data bit width to ensure that the receiving speed matches the output and prevent “data overflow.”

2. Signal Processing: Complete Format Conversion

Start different modules based on signal type:

· Analog signalThe A/D conversion chip (such as ADI’s AD9288) converts analog pixels into digital signals. The filtering circuit eliminates electromagnetic interference, prevents “snow” in the image, and avoids “burrs” at the edges of components from affecting measurements.

· Digital signalThe FPGA chip (such as the Xilinx Artix series) parses the protocol, extracts pixels and synchronization signals, and stores them in a buffer. If the data is compressed, it first decompresses the data to ensure that the original image is transmitted.

3. Data transmission: Ensuring real-time stability

The processed data is transmitted to the computer via interfaces such as PCIe and USB 3.0:

· High-speed interface priorityThe industrial-grade PCIe interface offers a bandwidth of up to 8 GB/s via PCIe x4, sufficient for real-time data transmission from 20-megapixel cameras running at 60 fps. The USB3.0 interface provides a bandwidth of approximately 5 GB/s, making it suitable for scenarios requiring frame rates below 30 fps.

· Cache anti-stutteringBuilt-in 1GB–4GB DDR3/DDR4 cache stores 1–2 frames first before transferring them in batches, preventing data “stream interruption” when the computer is handling other tasks.

III. Key Parameters: 3 Indicators for Model Selection

When choosing a capture card, focusing solely on the “interface” can easily lead to buying the wrong one or wasting money. By paying close attention to these three key indicators, you can precisely match your needs:

1. Number of channels: Determines the number of cameras that can be connected simultaneously.

The number of channels refers to the number of camera signals that can be processed simultaneously, and it is categorized as single / dual / quad channels:

· Single channelSuitable for single-camera scenarios (such as single-part defect detection), offering high cost performance.

· Multi-channelFor multi-camera collaboration (such as PCB panoramic inspection), a four-channel acquisition card can synchronously receive data from four cameras, increasing efficiency by a factor of four.

Note: Multi-channel systems must support “synchronous triggering” to prevent image misalignment caused by timing errors between channels.

2. Sampling frequency: Determines processing speed

The sampling frequency (in MHz) refers to the number of pixels processed per second and directly determines the maximum frame rate supported by the camera. The formula is:Sampling frequency ≥ Camera resolution × Frame rate.

For a 20-megapixel, 60fps camera, the sampling frequency must be at least 120 MHz. Choosing a 100-MHz card would result in “insufficient processing capacity,” leading to frame drops. The maximum allowable frequency varies depending on the interface: Camera Link can reach up to 850 MHz, while GigE is around 200 MHz. Therefore, when selecting a card, it’s advisable to leave a margin of 10% to 20%.

3. Resolution and Precision: Factors Affecting Image Clarity

· Resolution supportIt must be compatible with the camera’s maximum resolution. If the camera has 50 million pixels but the capture card only supports 20 million pixels, the image will be compressed, resulting in loss of detail and affecting high-precision inspections such as chip pin detection.

· A/D conversion accuracy (analog camera)The higher the precision, the more grayscale levels there are. An 8-bit (256 levels) is suitable for scenarios with significant color differences, while a 12-bit (4096 levels) can capture subtle brightness variations and is ideal for detecting scratches on transparent materials.

IV. Pitfall Guide: Distinguish Between “Frames” and “Fields”

In simulated or interlaced scanning camera scenarios, the terms “frame” and “field” are easily confused, which can affect image acquisition efficiency.

· Frame: A complete frame is equivalent to one photograph; the frame rate refers to the number of complete photographs captured per second. A progressive-scan camera outputs one frame at a time, making it suitable for detecting static objects or slowly moving subjects.

· fieldFor “half of a frame,” interlaced cameras enhance “smoothness” by dividing one frame into odd fields (rows 1, 3, 5…) and even fields (rows 2, 4, 6…), transmitting them separately, and then recombining them into a complete frame after transmission.

Avoid pitfalls: When detecting high-speed objects, prioritize line-scan cameras + acquisition cards; if using interlaced cameras, choose an acquisition card that supports field synthesis to prevent "image half missing."

5. Scene Adaptation: Select Based on Needs

There is no “universal” image acquisition card; you need to match it according to the specific application scenario.

· High-precision applications (semiconductor inspection)High-resolution, multi-camera synchronization required; select a multi-channel card with Camera Link/CoaXPress interface, featuring a sampling frequency ≥200 MHz and A/D resolution ≥12 bits.

· High-speed production line (beverage bottle inspection)For high frame rates and real-time transmission, choose a PCIe single- or dual-channel card with a sampling frequency ≥300 MHz that supports hardware triggering.

· Low-cost scenarios (teaching, small-batch testing)High cost performance and easy installation are required; choose a single-channel USB3.0 card with a sampling frequency ≥100 MHz, compatible with common USB cameras.

Conclusion: The image acquisition card is the “invisible cornerstone” of machine vision.

In a machine vision system, the camera is the “eye,” the algorithm is the “brain,” and the capture card is the “central nervous system.” Without these components, even the highest-resolution camera and the most sophisticated algorithm would be utterly “useless.”

The core logic for selection: First, clearly define the application requirements (number of cameras, frame rate, and accuracy); then, match these with the appropriate number of channels, sampling frequency, and resolution. Only by choosing the right acquisition card can a machine vision system fully realize its value—delivering “precise detection and efficient production”—and become an “quality guardian” in industrial automation.