Basic rules for a successful facial biometrics project

Regardless of whether you plan to transfer the video surveillance system already available at the enterprise to biometric identification, or create “smart” video surveillance from scratch, the result will directly depend on the fulfillment of a number of technical requirements.
In the RecFaces blog, we have already described how an incorrectly installed camera or an unsatisfactory video stream quality can affect the performance of even the most reliable facial recognition algorithm. Today we will talk about the basic rules for setting up and installing cameras, as well as other nuances that must be considered for the successful implementation of a biometric video surveillance project.
Contents
- Requirements for cameras
- Requirements for the installation of video cameras
- Calculating the number of cameras
- Calculating the storage size for an archive
- Check-list for preparation for the implementation of the biometric video surveillance project
Requirements for cameras
The Id-Guard software solution from RecFaces has no brand restrictions for manufacturers of video surveillance cameras. However, for the effective operation of the facial recognition algorithm, the equipment must meet a number of characteristics. The main task is to make the transmitted image as clear as possible. The higher the quality of the video stream, the more accurate the identification result will be.
Camera resolution – preference should be given to FullHD. From 1080p (1920×1080) and higher.
The frame rate of the video stream – at least 15 fps (frames per second). Depending on the project goals, the video stream from the cameras can either be directly broadcast in the Id-Guard system or registered separately in it. But in any case, the higher this parameter, the more accurate the picture is.
The focal length and width of the lens – in most cases, f 14-50mm is sufficient. These parameters play a special role in projects where the specified height of the camera installation does not allow achieving sufficient quality of “capturing” faces. Zoom settings help to increase the size of the face in the frame, which makes the biometric profile of a person more “readable”.
The camera focus should be set so that the face in the frame is clear, high-contrast and clearly visible to the naked eye.
Pixel density is the recommended value of at least 1000 pixels per meter. If necessary, you can use a lower value, but this may negatively affect the accuracy of identification.
Recognition angles – from the point of view of biometric identification, the ideal position of a person’s head is a full-face with a look directed exactly at the camera. It is recommended to take such photos when creating a database of profiles of employees of the enterprise. But in reality, passing by the cameras, people do not always keep their heads straight. The facial recognition algorithm is able to identify a person even if the angle of rotation of his head is 90 degrees. However, to get the desired results, it is necessary to optimize the settings of both the cameras themselves and the system.
For cameras, the recommended angle of rotation of the head from the front image is ± 15 degrees, the angle of deviation of the head is ± 15 degrees, and the angle of inclination of the head is ± 10 degrees. These parameters can be supplemented by fine-tuning the Id-Guard system. Their optimization allows you to exclude unnecessary or low-quality photos from processing.
Requirements for the installation of video cameras
Errors during the installation of video cameras can negate the equipment settings and reduce the accuracy of the results of the biometric identification algorithm.
The camera position is always in the direction of the flow of people. The person must move along the line of sight of the camera with a minimum deviation or tilt of the head from the line of observation.
Height – the optimal height for installing a video surveillance camera is from 1.5 to 2.2 meters.
Distance – ideally, the identification zone of a person should start at 8-8.5 meters from the camera. This distance will allow you to “capture” more faces and take more pictures of individual people while they are in the observation area, which will increase the probability of correct identification.
Camera angle – an incorrectly determined camera angle can create so-called “dead zones” of observation even in a small room. If the horizontal distortion of the view is leveled by the focal length settings, but vertically – only with the help of a correctly calculated angle of inclination of the equipment.
Illumination level – for the correct operation of biometric identification systems for the installation of cameras, you should choose rooms with a uniform level of illumination without shadows or glare. The minimum illumination level is 150 lux, the maximum is 600 lux. The optimal value is 300 lux.
Programs for designing video surveillance systems – to determine the location of cameras, it is recommended to use special programs for creating a video surveillance system design. For example, IP Video System Design Tool. With this tool, you can determine the optimal points for installing cameras and calculate their tilt angles. And also to understand how the zones of identification, recognition and detection of faces will look in a particular room, which is important for the operation of biometric algorithms.
Calculating the number of cameras
The exact number of cameras depends on two key factors: the type of site and the business objectives of the project. For simple rapid detection of violators, you can use cameras at the entrance and exit. But if there are areas in the room that require special protection or a level of access control, you will not be able to do with two cameras. Additional equipment will also be required if the tasks of the video surveillance system include monitoring the movements of people on the territory of the object. The more cameras there are, the more accurately you can determine the trajectory of each person.
Calculating the storage size for an archive
Unlike conventional video surveillance archives, Id-Guard data requires significantly less disk space. The system stores all information about people’s passes in the photo format, to which a biometric profile from the database is linked. When investigating incidents, this data can be sorted by date, time and observation zone. 50 million visitor passes take up only 1 TB of memory. When calculating the exact amount of disk space, the following factors must be taken into account:
- The number of cameras that are planned to be installed at the facility.
- Average number of visitor passes per minute.
- The number of days of storing operational data (recent passes).
- Number of days for long-term data storage.
- The period of operation of the video surveillance system: around the clock or in the mode of the schedule of the working day of the enterprise.
Check-list for preparation for the implementation of the biometric video surveillance project
- Make sure that the main characteristics of the cameras (resolution, video stream frequency, lens and focal length, pixel density, etc.) meet the system requirements.
- Make sure that the planned locations for the installation of cameras meet the optimal ratio in height, distance and angle of inclination.
- Choose well-illuminated rooms.
- Decide on the tasks of the project: to prevent unwanted persons from entering the territory, to control the movement of employees around the facility, to protect specific zones or something else? The answers to these questions are the basis for calculating the number of cameras that you will need.
- Evaluate the parameters of the room. Whether there are zones with an increased level of access control or open spaces that will require the installation of additional cameras.
- Determine the amount of disk space required to store the data archive.
- Special programs will help to automate the process of building a system design and determine the amount of equipment. You can also resort to the advice of specialists.
You can learn more about the rules for installing and configuring equipment, as well as about the operation of biometric identification algorithms in a video surveillance system at the RecFaces free webinars.