- Created by Aleksandr Savkin, last modified by Vasily Selivantsev on Dec 28, 2023
Recommendations for camera selection and setup
To work with the module, it is recommended to use a camera with a varifocal lens, which will allow you to change the viewing angle, zoom in or remove the shooting area without changing the position of the camera.
It is often the case that customers do not have ideal conditions for camera placement, as cameras have to be installed where permitted or where installation is possible, and then the image quality is adjusted through camera settings.
We do not recommend using cameras with fisheye lenses.
The size of the image sensor must be at least 1/2.8"
The aperture of the lens should be no worse than F1.6 (the lower the number after F, the better the aperture)
If there are high-contrast areas with different degrees of illumination in the shooting area, it is recommended to use cameras with a hardware WDR
The camera must be set to the lowest possible shutter speed. Values greater than 1/100 of a second are not allowed. Recommended values: 1/200 or less. If the picture darkens or noise appears at the recommended shutter speed values, you should replace the camera with a model with the best sensor sensitivity characteristics or increase the illumination of the shooting area.
In the camera's stream settings, the lowest possible GOP value and the minimum stream compression should be selected. It is recommended to choose a constant bitrate type with values corresponding to the maximum image quality.
The depth of field should, if possible, cover the entire area of the intended detection. Sharpness should be tuned to get the highest quality picture in the area of the best conditions for face recognition.
It is recommended that features such as noise reduction and other digital image conversions be disabled.
Visually, the picture should be clear, without distortion. Faces in the image should be sufficiently contrasting, illuminated and clearly visible to the naked eye.
In the resulting image, the distance between the pupils should be at least 60px.
It is recommended to use a camera with any resolution, but such that the size of the face in the frame is greater than 128px.
Recommendations for choosing camera installation angles and room lighting.
- There should be good lighting in the room where face recognition is performed.
For high-quality operation of the module, the camera should be installed in places with a centralized and structured passage area, for example, the front door to the building. Wide halls with chaotic movement of people are a bad place to install a camera.
Do not try to block a wide area with one camera. In this case, it is recommended to install multiple cameras.
The direction of the shooting should be such that people's faces look directly into the camera lens. A slight rotation of the camera horizontally or vertically is allowed, but not more than 30 degrees.
The best recognition quality is achieved when the faces are deflected by no more than 15 degrees.
Which errors are most common?
One of the frequent installation options that allows you to choose an area of the most structured flow of the minimum width is the installation of a camera opposite the entrance to the building / room.
On a bright sunny day, this can lead to strong backlights, and, as a result, a deterioration in image quality.
To combat this effect, you should use cameras with hardware WDR, or provide sufficiently powerful artificial lighting of the persons being filmed.
With a high ceiling or in pursuit of a minimum vertical angle of deflection, the camera has to be lowered from the ceiling on a long bracket, which can lead to strong fluctuations, for example, from air currents in the room.
Even slight fluctuations can introduce strong blurs into the image. In this case, it is recommended to apply additional measures for fastening (for example, additional supports / stretches).
Recommendations for photos used for recognition
For recognition, TRASSIR uses all photos in the Face Database. The probability of recognizing a person who has fallen into the camera lens depends on the quality of the photos uploaded to TRASSIR. To increase the probability of recognition, use the following recommendations:
For one person, you can upload several photos, one of which must be taken in full face, and the rest are allowed to rotate no more than 30 degrees vertically or horizontally.
Photos in which a person is depicted in profile will significantly reduce the likelihood of recognition of this person.
If a person wears glasses, then to improve recognition, upload a photo with glasses.
Photos in which a person's face is blurred, highlighted or in the shadows will significantly reduce the likelihood of recognition.
The face in the photos must be at least 128x128px.
Example of photos that improve the quality of recognition
Example of photos that reduce the quality of recognition
- No labels