What does an AI-powered video camera consist of — Part 2: 

Lens in AI Vision Cameras.

A lens is an optical system that forms an image and projects it onto the surface of the image sensor.

A camera lens consists of several key elements (see picture above):
▪️ optical lens elements that focus light onto the sensor
▪️ aperture (iris) that controls the amount of light
▪️ zoom mechanism that adjusts the focal length and field of view
▪️ lens mount that provides a mechanical connection to the camera
▪️ housing that ensures precise alignment and protection of all components

The quality of the lens directly affects:
▪️ system resolution
▪️ accuracy of computer vision algorithms
▪️ geometric distortions
▪️ camera sensitivity
▪️ depth of field

AI Vision systems place high demands on optics because machine vision algorithms are sensitive to distortions, noise, and loss of detail. To select the correct lens, it is necessary to understand the main parameters:
▪️ Focal Length
▪️ Iris, Aperture, F-number
▪️ Sensor Format (sensor size)
▪️ Lens mount types
▪️ Lens resolution
▪️ Distortion
▪️ Depth of Field
▪️ IR correction

Focal length determines the field of view of the camera, measured in millimeters and is usually denoted by the letter f .

Depending on the ability to change focal length, lenses can be:

▪️ Fixed (prime lenses) - Have a single, constant focal length.

▪️ Varifocal lenses - Allow manual adjustment of the focal length.

▪️ Motorized (zoom lenses) -Varifocal lenses with an electric motor that allows remote adjustment of the focal length.

The formula for calculating the field of view angle for an ideal lens is:

Also easily calculate the width and height of the scene using the following formulas:

▪️ Telephoto lenses (10–45° diagonal FOV)
Lenses with a large focal length that provide a narrow field of view and allow magnification of distant objects.

▪️ Normal lenses (45–60° diagonal FOV)
Close to the way the human eye perceives space, providing a natural perspective without noticeable distortion.

▪️ Wide-angle lenses (60–90° diagonal FOV)
Provide a wide field of view, allowing the camera to capture a larger area of the scene.

▪️ Ultra-wide lenses / Fish-eye (up to 180° diagonal FOV)
Provide an extremely wide field of view and create strong barrel distortion.

The lens speed (light transmission) indicates how much the light flux is reduced after entering the lens. This parameter is affected by the diameter of the aperture (Iris) opening, the quality of the lenses (transparency), and several other optical characteristics.

The aperture number, or F-number, is the result of dividing the focal length of the lens (f, mm) by the diameter of the effective aperture (Iris) opening.

The smaller the F-number, the better the lens collects light, meaning that more light reaches the image sensor. Under low-light conditions, a lens with a smaller F-number usually produces a higher-quality image. As the F-number increases, the depth of field increases. Typically, a lens with a low F-number is more expensive than a lens with a higher F-number.

The iris can be fixed or adjustable. In the case of an adjustable iris, it can be controlled automatically or manually.

A more modern standard for precise automatic aperture control. The system uses a P-Iris lens equipped with an electric actuator and specialized software. The system compensates for the limitations of traditional auto iris systems.

It is important to select the correct lens size for a camera sensor. Each lens is designed for a specific sensor format.
▪️ If a lens is designed for a smaller sensor, vignetting occurs — the image becomes darker at the corners (see left illustration).
▪️ If a lens is designed for a larger sensor, the effective field of view becomes smaller, and part of the image falls outside the sensor and is lost (see right illustration). In this case, the image appears zoomed-in, and focusing issues may occur.

If you cannot find a suitable lens for your sensor format, you can use a lens designed for a larger format. However, you will need to use Extension Tube (distance rings) to increase the distance between the lens and the sensor.

This is required to: reduce the zoom effect, achieve correct focus
Extension Tube are available in different thicknesses, from 0.5mm, 2 mm, 5 mm, 8 mm, 10 mm to 50mm
They can be combined to achieve the desired image size and focus.

There are three main lens mount standards:
▪️ S-mount (M12)
▪️ C-mount
▪️ CS-mount

The S-mount (M12) standard is used in compact and mini dome cameras.

Features
▪️ M12 thread
▪️ compact size
▪️ low cost
▪️ widely used in IP cameras and embedded vision

Disadvantages
▪️ limited optical quality
▪️ strong distortion at wide angles

It's standard for industrial machine vision.
C-mount and CS-mount mounts look identical and both have a 1-inch thread with 32 threads per inch (TPI).
The CS-mount is an improved version of the C-mount standard.
This standard supports large sensors: 4/3", 1", 1/1.8", 1/2.8", 1/3".

The difference between these mounts lies in the distance from the lens to the image sensor:
▪️ C-mount = 17.526 mm
▪️ CS-mount = 12.5 mm

Due to the identical thread, the lenses are mechanically compatible but not optically compatible.
A C-mount lens can be used with a CS-mount camera by using a 5 mm tube (C-to-CS adapter), which is typically included with the camera. In the opposite direction, the lenses are not compatible.

Advantages
▪️ high optical precision
▪️ low distortion
▪️ support for large sensors

Disadvantages
▪️ higher cost
▪️ larger size

Lens resolution is the ability of the optics to transfer fine image details to the sensor. If the lens has insufficient resolution, it becomes a bottleneck of the entire system, and the sensor cannot utilize its full number of pixels.

As a result:
▪️ the image appears soft
▪️ fine details are lost
▪️ the accuracy of computer vision algorithms decreases

The lens must match the resolution of the sensor. The smaller the pixel size, the higher the requirements for the lens.
If the lens is insufficient, optical blur occurs, and the sensor cannot achieve its actual resolution. Therefore, for “megapixel” cameras, lenses with resolution equal to or higher than the sensor resolution must be used.

Lens resolution is measured in line pairs per millimeter (lp/mm). A line pair consists of a pair: one black line and one white line. If a lens can resolve 100 lp/mm, it means it can distinguish 200 individual lines per 1 mm. To fully utilize the sensor capabilities, the lens resolution must correspond to the pixel size.

The maximum spatial frequency is determined by the Nyquist criterion, where pixel size is expressed in millimeters:

Sony IMX415, pixel size is 1.45 µm = 0.00145 mm, substituting:

Most wide-angle lenses have distortion — distortion (from Latin distorsio, distortio — curvature).

Distortion is an aberration of optical systems in which the linear magnification factor changes as the imaged objects move away from the optical axis. As a result, the geometric similarity between the object and its image is violated.

Barrel distortion — the image appears “convex”.
Barrel distortion occurs when the actual field of view is wider than the calculated one.

Pincushion distortion — the image appears “concave”.
Pincushion distortion occurs when the actual field of view is narrower than the calculated one.

For AI Vision, this is critical because:
▪️ scene geometry is distorted
▪️ detection accuracy decreases

Usually, distortion is reduced using camera calibration and distortion correction.

Sony IMX415, pixel size is 1.45 µm = 0.00145 mm, substituting:

Depth of field defines the range of distances within which the image remains sharp, clear, and not blurred. Depth of field is the distance in front of and behind the focus point within which objects appear sharp. Depth of field is important, for example, in parking monitoring, where it may be necessary to read license plates at distances of 20, 30, and 50 meters.

It depends on: focal length, aperture, distance to the object, sensor size.

A large focal length, a wide aperture, or a small distance between the camera and the object limit the depth of field. For machine vision, maximum depth of field is often required so that objects at different distances remain sharp.