As the name suggests, a linear array camera captures images in a linear manner. Its sensor is arranged in a linear configuration.

For example, a front-facing camera with a resolution of 640×480 means it has 640 pixels horizontally and 480 pixels vertically.

The resolution of a linear array camera is defined solely in the horizontal direction; for instance, a 2048-pixel linear array camera has 2048 pixels horizontally and typically 1 pixel vertically (excluding RGB and TDI cameras).

Regarding the sensors of linear array cameras

During the 1970s, MOS sensors were predominantly used. By the late 1970s, CCD technology began to rapidly advance and has remained mainstream ever since. CMOS sensors emerged around the mid-1980s; however, due to technological limitations, CCDs offered slower image capture speeds than CMOS sensors, and until 2010, CMOS sensors were more expensive. After 2010, major camera manufacturers vigorously developed CMOS cameras, which have since found widespread practical applications.

In my opinion, CMOS sensors will become the mainstream choice for future linear array cameras. (You can research the advantages and disadvantages of these two sensor types online; the key differences lie in imaging speed and sensitivity.)

Key parameters of a linear array camera:

Resolution: Number of pixels, i.e., the number of pixels on the sensor.

MAX DATA RATE (should be called camera clock): refers to the maximum amount of data the camera can process per second

Linerate frequency: refers to the maximum number of image rows the camera can capture per second.

For example, with a resolution of 8192×1 and a data rate of 160 MHz, the camera's line rate is calculated as 160 MHz ÷ 8192 = 19,000 lines per second.

The system can capture up to 19,000 rows per second, with a horizontal resolution of 8,192 pixels and a vertical resolution of 19,000 pixels. The image obtained in one second has a size of approximately 160 MB.

Another critical factor is the pixel size and lens dimensions. Typically, CCD pixels are at least 5 μm in size; any smaller would be impractical to manufacture, and their sensitivity would be significantly lower. In contrast, CMOS pixels can be nearly half the size of CCD pixels.

The choice of camera is crucial, as it directly impacts the overall equipment cost: higher pixel counts require larger lenses, greater data volumes necessitate data cables with higher transmission rates, and additional image processing cards are required. Moreover, large data volumes impose stringent computational demands on both hardware and software systems.

Taking modern mainstream CCD cameras as an example, due to their limited image sampling rate—typically each tap can capture up to 60 MHz of data—most cameras now employ a multi-tap processing approach, with each tap generally capturing 40 MHz. For instance, a 160 MHz camera has four taps, each capturing 40 MHz; thus, 40 MHz × 4 = 160 MHz. There are also variations such as single-tap (1), dual-tap (2), triple-tap (3), and octal-tap (8). Currently, CCDs have sampling rates below 400 MHz, whereas CMOS sensors can achieve up to 1.6 GHz (potentially higher in the future).

Cameras offer various output formats—8-bit, 10-bit, and 12-bit—but my primary focus is on 8-bit black-and-white images with a 256-level grayscale scale.

Available with single output: 1, 2, 3, ~8192 during imaging; dual outputs: 1, 3, 5, 7/2, 4, 6, 8; or 1, 3, 5–4095/4097,4099–8191.

You can get a general idea of the output format here (using the default values generally won't affect image capture).

The primary interfaces for linear array cameras remain GiGe and Cameralink as the mainstream standards, while high-speed cameras require HSLink.

The camera's primary settings include exposure, gain, and internal/external trigger modes, along with less frequently used options such as average image grayscale and offset settings.

Exposure: This setting is directly related to the camera's line frequency and must be lower than the maximum allowable line frequency.

Take the aforementioned 19,000-line camera as an example: to achieve this resolution, set the time interval to 1 second divided by 19,000 = 53 μs. However, there is additional delay involved; setting it around 47 μs ensures maximum line capture frequency. (The explanation needs clarification.)

The lower this value is, the darker the image will be; the higher it is, the brighter.

Gain/Offset adjusts the image's grayscale level and is useful when lighting conditions are poor. However, it reduces image contrast, which may affect analysis accuracy; therefore, we do not recommend its use. Even if employed, the adjustment should not exceed 20% of the default value.

For additional settings, please refer to the camera's user manual.

Line array cameras are primarily used in continuous production lines (web-based systems), such as those in steel metallurgy, non-ferrous metals, electronic materials, textiles, papermaking, and LCD manufacturing. In fact, any application suitable for surface array cameras can also be handled by line array cameras—though the cost remains a key consideration.

Let me illustrate with an example—first, consider a monocular camera case.

Surface Defect Detection Equipment for Electronic Copper Strips

The electronic copper strip has a width of 450 mm and a production line speed of 120 meters per minute; the minimum defect size required for inspection is 0.2 mm.

When selecting a camera model, consider a 4096-pixel linear array sensor. With a standard F-stop lens, its horizontal resolution reaches approximately 0.11 mm, enabling detection of defects as small as 0.2 mm. For vertical resolution selection: 120 meters per minute equals 2 meters per second, or 2000 mm per second.

To achieve uniform horizontal and vertical resolutions, the resolution should be calculated as 2000 mm / 0.11 = 18,180 lines; the camera must capture 18,180 lines per second to obtain a full-frame image of the product. Therefore, we can select a camera with 4,096 pixels and a line frequency of 19,000. Such cameras are capable of capturing full-frame images of products.

Just now, a group member named Shanghai-Alex-VC asked whether it's possible for all scans within one second to fall on the same row.

It won't work with external triggering.

External triggering means that an external pulse is applied to the camera, causing it to scan one row. The production line speed determines the scanning frequency—higher speeds result in higher frequencies, while lower speeds yield lower frequencies.

External triggering is primarily implemented using encoders (though this digresces slightly into encoder specifics). There are two main types of encoders: fixed-pulse types (e.g., 1 mm equals 1 pulse, constant value) and revolution-based fixed-pulse types. Encoder output signals vary widely, including formats like LineDiver; similarly, cameras capture signals in multiple formats such as TTL, LVDS, or differential signals.

Take electronic copper strips as another example: here we use a fixed pulse encoder that generates one pulse per 1 mm, meaning the camera scans and captures an image every 1 mm. This results in a horizontal resolution of 0.11 mm and a vertical resolution of 1 mm, leading to significant image distortion. Two solutions are available: first, select an encoder with a frequency matching the line frequency; second, adjust the line frequency through the camera settings.

All cameras feature a convert mode, which converts received pulses into required values. For instance, setting both horizontal and vertical resolutions to 0.11 mm simply requires a ninefold increase in resolution. Each pulse triggers nine automatic consecutive scans, resulting in a line resolution of 1 mm / 9 = 0.11 mm—ensuring uniform resolution across both dimensions.

For example, when a 1 mm area generates one pulse, the camera captures that pulse and scans nine rows:

When the production line operates at 1 meter per second, the camera scans 1000 mm × 1 × 9 = 9000 lines per second. For every meter traveled by the production line, the encoder generates 1000 pulses, and the camera captures 9000 lines. At 2 meters per second, the camera scans 2000 mm × 1 × 9 = 18000 lines per second; for every 2-meter increment, the encoder generates 2000 pulses, and the camera captures 18000 lines. With proper external triggering settings, continuous scanning on the same line will not occur.

This brings us back to color cameras, which operate on a three-line scanning principle: the three lines are scanned simultaneously in a single row, and their combined output forms the final color image.

There are also TDI cameras available in 8-,16-, and 32-line configurations, with the highest capacity reaching approximately 512 lines. These cameras capture images within a single line to achieve optimal quality; they are expensive, require minimal lighting conditions, but demand very high data rates.

For inquiries regarding the selection of line cameras, lenses, and light sources, please contact us to discuss setting up and choosing a line scanning system.

With the widespread adoption of machine vision and the increasing speed and precision of industrial assembly lines, line scanning systems have gained growing recognition among both vision engineers and end-users.

First, I will provide a general overview of the line scanning system. This system is used when there is relative motion between the object being measured and the camera. The line scanning camera captures data at high speed; after completing the capture of one line, it moves precisely to the next unit length and proceeds to capture the subsequent line. Over time, these segments are combined to form a two-dimensional image, similar to that captured by a surface-array camera, with the key difference being that the image can extend infinitely in height. Subsequently, software crops this "infinitely long" image into a picture of a specified height for real-time processing or storage in a buffer for later processing.

The visual system comprises a line-scanning camera, lens, light source, image acquisition card, and visual software.

The motion control section includes a motor, motor driver, motion control card, or PLC; an encoder may also be required to ensure synchronization between the captured images and the conveyor belt.

Due to the large amount of data required for linear scanning, a high-performance industrial computer is necessary, equipped with ample memory and a hard drive, and featuring motherboard slots supporting PCI, PCI-E, or PCI-X interfaces.

Generally, the configuration selection for a surface array vision system follows this sequence:

Camera + Acquisition Card → Lens → Light Source

The same principle applies to line-array projects: the resolution and row scanning speed of the line-array CCD camera are determined based on the system's detection accuracy and speed requirements, along with the appropriate acquisition card. However, when selecting the camera lens mount, the lens type must also be considered, followed by final selection of the light source.

Selection of Line Array Cameras (Line Array Industrial Cameras)

Calculate resolution: divide the width by the minimum detection accuracy to obtain the number of pixels per row

Select Camera: Divide the width by the number of pixels to obtain actual detection accuracy

Divide the movement speed per second by the precision to obtain the number of scan lines per second.

Select the camera based on the above values

With a width of 1600 mm, precision of 1 mm, and movement speed of 22000 mm/s

Camera: 1600/1 = 1600 pixels

Minimum 2000 pixels; selected as a 2K camera

1600/2048 = 0.8; actual accuracy

22000mm/0.8mm＝27.5KHz

Select a camera with 2048 pixels and 28 kHz sampling rate.

Selection of Linear Array Lenses

Why should you consider lens selection when choosing a camera? Common linear array cameras offer resolutions of 1K,2K,4K,6K,7K,8K, and 12K, with pixel sizes ranging from 5μm to 14μm. Consequently, chip dimensions vary from 10.240 mm (1K×10μm) to 86.016 mm (12K×7μm). Clearly, the C-mount interface falls short, as it only supports chips up to 22 mm (approximately 1.3 inches). Many cameras use interfaces such as F, M42X1, or M72X0.75; each lens interface corresponds to a specific flange distance, which determines the lens's working distance.

Optical magnification (β, Magnification)

Once the camera resolution and pixel size are determined, the sensor size can be calculated. Dividing the sensor size by the field of view (FOV) yields the optical magnification: β = CCD / FOV.

Interface (Mount):

The main models include C, M42x1, F, T2, Leica, and M72x0.75; once identified, the corresponding lens length can be determined.

Flange Distance

Back focal length refers to the distance from the camera interface plane to the sensor chip and is a critical parameter determined by the camera manufacturer based on their optical design. Different manufacturers may produce cameras with varying back focal lengths, even when sharing identical interface specifications.

With the optical magnification, interface specifications, and back focal length known, the working distance and circle length can be calculated. After determining these parameters, another critical step is to verify whether the MTF value meets adequate standards. Many visual engineers lack understanding of MTF, yet for high-end lenses, it serves as a fundamental metric for evaluating optical performance. MTF encompasses comprehensive information—including contrast ratio, resolution, spatial frequency, and chromatic aberration—and provides detailed characterization of optical quality both at the lens center and along its edges. If not only the working distance and field of view meet requirements but also the edge contrast is suboptimal, reevaluation of selecting a lens with higher resolution becomes necessary.

Selection of Line Scanning Array Light Sources

In line scanning applications, commonly used light sources include LED sources, halogen lamps (or fiber-optic light sources), and high-frequency fluorescent lamps.

Halogen lamps, also known as fiber-optic light sources, offer exceptionally high brightness but have notable drawbacks—short lifespan of approximately 1,000–2,000 hours requiring frequent bulb replacement. These lamps utilize a halogen bulb with a specialized optical lens and beam splitter system, transmitting light via fiber optics with power output reaching up to 250 watts. Also referred to as cold light sources, they maintain a cool emission end and stable color temperature during fiber transmission, making them ideal for applications sensitive to ambient temperature, such as illumination in two-dimensional measurement instruments. Line-scanning halogen lamps often incorporate glass condenser lenses at the emission port to enhance focalization and luminosity. For longer linear light sources, multiple halogen units may be employed simultaneously to illuminate a single fiber optic cable.

High-frequency fluorescent lamps operate on a principle similar to that of fluorescent tubes, with the key difference being that their tubes are industrial-grade components. They are characterized by suitability for large-area illumination and high brightness.

The cost is low, but the major drawback of fluorescent lamps is their flickering and rapid light decay. Fluorescent lamps require a high-frequency power supply, meaning the light source's flicker frequency must be significantly higher than the camera's image acquisition rate (for line-scanning cameras, this corresponds to the line scan frequency) to eliminate image flicker. A dedicated high-frequency power supply can achieve a frequency of up to 60 kHz.

The LED light source is currently the mainstream choice for machine vision applications, characterized by long lifespan, excellent stability, and extremely low power consumption.

1. DC power supply with no flicker.

2. Professional LED light sources have an exceptionally long lifespan (for example, those from American AI brands maintain at least 50% brightness for 50,000 hours).

3. The brightness is also exceptionally high, approaching that of halogen lamps and continuously improving with advancements in LED manufacturing technology (the current brightness of American AI line light sources reaches up to 90,000 LUX).

4. It can be flexibly designed into various types of linear light sources, such as direct illumination, those with a condensing lens, backlight configurations, coaxial arrangements, and bowl-shaped diffuse reflection configurations.

5. Multiple colors are available, including red, green, blue, white, as well as infrared and ultraviolet. Depending on the surface characteristics and material properties of different subjects, selecting appropriate colors (i.e., light sources with different wavelengths) ensures optimal image quality.

Analysis of the angle between the line scan camera, light source, and the object being measured

Taking glass inspection as an example, the defects that need to be detected include: dirt spots, stones, impurities, bubbles, scratches, cracks, and fractures. These defects can generally be divided into two categories: those on the glass surface and those inside the glass. Different defects appear with varying shades of gray in images—ranging from black and white to gray tones—and their contrast varies depending on the illumination angle or camera viewing angle. For instance, at certain angles, one type of defect may exhibit optimal contrast, while others may appear less distinct or even completely invisible. This necessitates extensive analysis and comparison to determine the optimal light source configuration and the relative angles between the camera, light source, and the object under inspection. As shown in the figure below, the camera and light source are installed at different angles for separate testing.

the result shows that ：

Stains are easily noticeable under either front lighting or backlighting.

For stones and impurities, illumination from the front perpendicular to the normal surface or backside penetration illumination is required.

The bubbles have irregular shapes, and their formation causes as well as direction must be analyzed; back lighting should be employed.

Scratches and damages; cracks are easily visible under low-angle front lighting, requiring back-side illumination.

Moreover, the aforementioned deficiencies are not independent but interact with each other. The statistical analysis is presented as follows.

Considering all factors, the final solution combines oblique backlighting with front illumination, and the camera is mounted perpendicular to the normal direction.

Adjustment of Light Source and Lens

In a line scanning system, the effective working area for both the light source and camera is a narrow strip. This requires ensuring that the light source illuminates this brightest narrow strip and remains completely parallel to the camera sensor; otherwise, only a single bright spot at the intersection can be captured. Consequently, mechanical installation and calibration are quite labor-intensive. Additionally, due to the relatively wide beam width, two specific requirements apply to linear light sources: uniformity and straightness. Variations in brightness across different positions of a linear light source directly affect image luminance, a characteristic that LEDs exhibit better control over than halogen lamps. The straightness of the light output depends on the consistency of the LED's emission angle, the straightness of the focusing lens, and the straightness of the linear light source housing.

Given the complex on-site environment, clients often prefer to allocate additional time for field debugging. However, as previously discussed, factors such as camera configuration, lighting conditions, and the relative angles between the object under test and the measurement equipment can significantly impact testing accuracy. Therefore, we recommend conducting laboratory tests first; only after developing a comprehensive plan should field debugging proceed. This approach ensures maximum reliability and enhances debugging efficiency, as service provision itself constitutes a cost factor.

In addition to its mechanical structure, the main components of a line scanning system also include machine vision and motion control.

To ensure that the captured images are not stretched or compressed, you must adhere to the principle of equal horizontal and vertical resolution.

First, define the following variables:

1) Number of pixels per line in the linear array camera (unit: pixel): Hc

2) Width of the target object (unit: m): Lo

3) Operating speed of the target object (unit: m/s): Vo

4) Line array camera line scanning rate (unit: Hz, i.e., lines/s): Vc

5) Time for the target object to move in one frame of scanned image (unit: s): To

6) Scan time of a line array camera for one image frame (unit: s): Tc

like that ，

The horizontal resolution is: Lo/Hc, where Lo/Hc represents the value on the calibration paper divided by a single camera pixel (4096).

The vertical resolution is: (Vo* To) / (Vc* Tc), where Vc and Tc are speed values per line.

It is easy to see that To = Tc.

Based on the principle of "equal horizontal and vertical resolution," the formula is as follows:

Lo/ Hc = Vo / Vc

The camera's line scan rate is:

Vc = Hc * Vo / Lo