Neural networks for business needs: easy steps to implement biometrics
Cameras... Cameras are at every turn. They record moving violations, identify us in a crowd, allow us to pay using nothing but our face, and protect our digital identity.
How do face-recognition systems work? Which sectors of the economy could benefit the most from biometrics implementation? And how to ensure seamless system integration?
- How face-recognition technologies work
- Why biometric solutions come in handy
- How to choose the right IT-solution for integration
- Guidelines for system operation and camera installation
How face-recognition technologies work
Biometric technologies (and face recognition in particular) are gradually becoming an integral part of our life. The neural networks behind them are able to process visual data, identify and verify information about who exactly is in the frame. RecFaces’ experts have already dwelled on the subtleties of specialized terminology in the article: “Dot the i's and cross the t's: identification, verification and authentication in biometric products”. In layman's terms, smart biometric solutions are based on two technologies. The first one finds and captures faces from the video, then “crops” them according to a given silhouette and “aligns” (rotates or centers) them if necessary. Once the photo meets the necessary parameters, it is time for the second technology to take over. It analyzes the photo taken by the first neural network and compares it with the database.
Woody Bledsoe, a scientist from the University of Texas, is widely regarded as the father of biometrics. Way back in the '60s, long before the era of social media and selfies, Woody, together with other researchers, created a database of 800 pictures of people taken from various angles. It is noteworthy that his counterparts’ results were not so groundbreaking. Previously, algorithms used to be trained to tell symbols apart, and now they are able to identify faces. Most of the key elements that have given impetus to biometrics, such as databases or comparisons thereof, are all credit to Woody Bledsoe.
Why biometric solutions come in handy
Biometrics is increasingly coming into use in various fields. User identification, security, and cashless payment are just a few examples of its successful implementation. Technology and algorithms are getting more and more sophisticated and face recognition is becoming less of a hype and more of a common thing. Due to artificial intelligence lying at the core of biometrics, law enforcement agencies can systematize information more efficiently: automated databases speed up the process of tracking and catching criminals and store the information on their criminal record. Another high-tech area that has reaped the benefits of biometrics is certainly medicine. Thanks to cameras installed in retirement homes, it is possible to monitor the medication regimen and the residens’ condition. Cutting-edge genetic applications save tons of time and money by analyzing scans to reveal one’s predisposition to rare diseases. Biometric technology has proven effective in the private sector of the economy as well. Cameras in stores track shoplifters and other offenders, help to enforce labor discipline of staff, and analyze customers’ behavior to improve service in the future. Wherever the human brain gets tired and runs out of concentration, artificial intelligence steps in and gets a grip on routine tasks 24/7 performing them much more efficiently.
Here are just some of the areas where biometric solutions could save time and reduce the workload of employees:
- process automation;
- demand stimulation;
- access control;
- air service security;
- comprehensive banking security;
How to choose the right IT-solution for integration
The biometric technology market offers a whole variety of face recognition software. Let us take a look at some of the most common integration types:
- Customized solutions. These products are developed for the needs of a particular customer, taking into account the peculiarities of their business processes, which obviously takes a lot of time. Such inflexibility of customized solutions results in the inexpediency of their implementation. The time it takes to develop a complex “customized” product cannot be entirely justified since the demand for this innovation is very likely to decrease, and the product itself might become outdated;
- Adjusting the algorithm to perfectly meet the customer's requests. In practice, such a strategy also takes a good deal of time and does not guarantee secure results which will adversely affect the product’s commercial appeal. Furthermore, like in the previous case, the final product might fail to comply with up-to-date business needs of the customer;
- box solutions that are easy to use and fast to integrate.
The success of a project largely depends on the right choice of both the vendor and technology. The comprehensive market analysis allowed RecFaces experts to see the whole picture, consider all pros and cons of each approach to integration software and finally opt for ready-to-use biometric products that take just 20 minutes to install and stand out among other face recognition solutions. The only thing the customer needs to do is to fill the database with photos and connect image sources to the system which can be easily done with the help of import tools and simple step-by-step instructions. As a result, our customers get a ready-to-use face recognition system and neither additional coding nor involvement of subject matter experts is necessary.
“Our decision to focus on developing ready-made biometric products stemmed, above all, from the results obtained from our pilot projects. Having analyzed general market demands, the company systematized them according to different industries and customers’ needs and came up with a kind of ‘product portfolio’ and a list of requirements for ready-to-use solutions. RecFaces already has several ready-made products under the belt. They are based on artificial intelligence algorithms, but unlike them, they are comprehensive ready-to-use solutions that do not require years of debugging and fine-tuning. The company's future plans include expanding its line of software products for the corporate sector," RecFaces experts report.
How can a user benefit from RecFaces ready-made products? Let us comment on the most significant ones:
- easy installation that takes 20 minutes;
- easy photo import tools for a quick start;
- user-friendly interface;
- clear reports in faces and figures;
- ready-to-use adapters;
- mobile apps;
- detailed documentation;
- full compliance with the personal data protection policies of different countries;
- open API for flexible integration;
To showcase the usability of ready-to-use products, RecFaces experts openly share their expertise during free webinars on smart biometric products’ implementation as well as provide 3-month demo licenses to allow customers to evaluate the benefits of the solutions in real operating conditions.
Guidelines for system operation and camera installation
RecFaces’ products are utterly flexible in terms of integration, since they can work with various brands of surveillance cameras. To ensure the correct operation of the facial recognition algorithm, you need to select equipment that meets a number of characteristics. The identification accuracy and the final system operation depends, to a great extent, on the quality of video image.
First, it is necessary to calculate how many cameras will do for an accurate video identification process. For that end, the two key factors should be taken into consideration: the type of enterprise and its prime objectives in terms of video surveillance. Now, let us move on to the characteristics of the cameras:
- camera resolution — Full HD, from 1080p (1920×1080) and higher, video frame rate — at least 15 frames per second;
- focal length and lens width — in most cases, 14-50 mm will do (these parameters are critical in projects where the given height of the camera installation encumbers adequate face “capturing” , and the zoom settings help to upscale the size of the image in the frame, which makes the biometric parameters of the person's face more “readable” to the system);
- the camera focus should be adjusted in such a way that the faces in the frame are distinct, high-contrast and clearly visible for the operator responsible for monitoring;
- recognition angles: from the point of view of biometric identification, the ideal position of a person's head is a full-face with the head turned to the camera. It is recommended to take such photos when creating a database, but in reality, hardly do people control the position of their head when passing by the cameras. Smart facial recognition algorithms are able to identify a person, even if their head rotation is up to 90 degrees, however, to obtain the desired results, it is better to optimize the settings of both the cameras and the system. As for the cameras, the permissible angle of head rotation from the frontal image is ± 45°, the optimal one is up to 15°, and the permissible angle of head inclination is up to 35°, the optimal one is ± 15°.
Once you have chosen a camera, you may move on to its installation which is also quite a tricky process. Here are some of the essential details to bear in mind: the room where the cameras will be installed, its perimeter, and whether it is equipped with demo screens for displaying some media content. Improper installation of the cameras nullify the equipment settings and reduce the accuracy of the results of the biometric identification algorithm. Pay attention to the following criteria:
- camera position — it should be in line with the direction of the flow of people (a person should always be within the camera’s viewing angles and only the minimal deviation or tilt of the head from the line-of-sight angle is admissible);
- the height of the CCTV camera installation is 1.5–2.2 meters, the distance is 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 surveillance zone, which will increase the accuracy of identification);
- the wrong camera tilt angle will create “gray zones” even in a small room (horizontal view distortions can be compensated by adjusting the focal length, but vertical view distortions can be compensated only by using the correct equipment tilt angle);
- recommended lighting level — it is better to choose evenly lit rooms without any shadows, highlights and glares (to decide on the location of the cameras, it is recommended to use special programs for video surveillance system design);
- extra objects in frame — ideally, there should be no interior elements in the camera frame that can somehow overlap or distort the image (screens displaying video content, glass or other obstacles, steam generators and even various fans).
To sum it up, when choosing and installing video surveillance systems, it is necessary to bear in mind the following things:
- formulate the principal objectives of the project: to prevent access of unwanted persons to the territory, to control employees’ getting about the territory of the company, to tighten control over some separate areas — the answers to these questions allow you to calculate the required number of cameras;
- take into account the parameters of the room: whether there are areas with additional access control or open space zones, that require more cameras;
- fix upon the store capacity of the disk necessary to store data archive;
- make sure the room where cameras will be installed meet all the requirements of optimal ratio of height, distance and angle, and they are well lit.
Learn more about facial recognition systems at RecFaces’ training courses. During our free webinars, RecFaces experts showcase the functionality and special features of our biometric products. You also have an opportunity to test the benefits of our box solutions in your own tech-infrastructure by getting a 3-month demo-license that we provide free of charge.
FAQ: biometrics and operation conditions of the system
Can RecFaces solutions detect age and gender of the visitors?
Can RecFaces solutions “read” people’s emotions?
If a visitor without a mask is identified, will pop-up notifications be generated?
Do masks, facial hair or colorful T-shirts mess with face recognition speed rate? Is the claimed similarity index the same for people wearing masks?
• if there is a high quality photo of a person in the database;
• how the person's face is placed in the camera frame;
• if the room is well-lit;
• what kind of medical mask a person is wearing.
Such attributes as masks, facial hair, hats and clothes have no effect on the speed of face recognition. For example, an Italian brand once tried to mislead AI algorithms by creating garments that are allegedly capable of fooling face recognition algorithms with the help of special ornaments. Our experts have studied such scenarios. Hacking biometrics, in general, is an extremely time-consuming task. The technology is constantly being improved and the algorithms are getting more and more sophisticated and now take into account more and more points on the face. For the purposes of face recognition, neural networks use those data that are less prone to distortion. Our experiments have once again proven that all these tricks do not work. Artificial Intelligence can detect and recognize all faces in the frame.
Are there any recommendations concerning camera configuration before adding it to the system? In which codec is the stream received via RTSP ?
• shutter speed, aperture, streams (stream setting: base frame, reference frame);
• H.264 codec;
• in some cases, white balance and brightness;
• FPS on the camera (if you want to reduce the load on the network);
• in some cases, white balance and brightness;
• focal length;
• FOV (field of view; sometimes it is better to set the camera in a corridor mode, for example, 9:16 or 3:4).
How many cameras can be connected to a video preprocessing server?
How does the system automatically synchronize profiles? Can lists be deleted in the same way they are added to the system? Are these folders stored where the system core is installed?
What are RecFaces' priority goals?
• flexible opportunities for forming a biometric profile of a person, including by adding several photos to it in order to increase identification efficiency;
• saving all detected persons’ photos in the archive without restrictions, which will significantly facilitate the search for necessary information in the future;
• service functions that are crucial for corporate systems: backup, recovery of critical data, monitoring or restarting system components and software, installation management on multiple servers.