Developing a Face Recognition System: How We’ve Done This and What We Have Learned Along the Way

It Is Important to Know Not Only What You Can Expect of a Well-Developed Face Recognition System, but Also How You Can Maximize Its Potential Through Its Optimal Installation and Correct Use.

A team that has spent decades helping others build software products, up until recently we had no product of our own. While many fellow-Ukrainians dream big and seek funding to turn their nearshore IT outfits into product companies, we just happened to follow quite a different scenario.

The mundane need to better secure our spacious office premises and the fact that we had several good AI experts with a fitting set of skills created a synergy that has influenced our company direction and made us into a product-driven software vendor. Moreover, we are now one with an innovative AI product a great many businesses would probably be feeling a lot more secure and confident with than without.

So, what can our product — an innovative Face Recognition system, — do, what makes it efficient, and how can one further hike up the efficiency of such a system by installing it more optimally on one’s office premises?

We’ve already mentioned in one of our previous articles that our proprietary Face Recognition system was completing development and we were adding several advanced features that in many contexts would be considered as no-frills. The bulk of the system has long been deployed and tested in a real-world setting in our office. Presently, the system is integrated with our HRM and card-enabled access control systems. It is used to secure access to two separate office doors. The number of visitors that our system needs to identify is continually on the rise (of late, we’ve been welcoming onboard at least 2 newcomers a week, and these guys had to be added to the database and issued with cards).

Hence, is the first tip we can give:

• Never pick a Face Recognition system that is not sufficiently configurable: your current business needs may change faster and a lot more drastically than you think. For instance, we’ll shortly have to be extending our system’s coverage to include an entire new office floor, but, in our case, this isn’t much of a concern or effort, as we, being unaware of this future expansion at the time, still took this possibility into consideration.

  Therefore, one must necessarily be able to quickly and easily add new doors, cameras, people and cards.

• The system you install must have a sufficient set of interfaces that will allow it to be used in conjunction with your access control and HRM systems for adding new people and entering other data.

Currently, the version of our system that is installed on our premises includes a total of 4 cameras.

One of the more important things the efficiency of your Face Recognition process will be heavily impacted by and definitely one you should pay a lot of attention to is the positioning of your cameras.

• Your cameras must be positioned so, that they will capture as many front-face images of your visitors, as possible. In other words, the front-face is always preferable to any other angle.

• Your cameras must be positioned in such a way, that your visitors' faces will be lit sufficiently.

The cameras send the visual data they collect to a server for the system to:

1. detect any possible movements.

2. locate a human face, if there is one.

3. identify the face, if there is one (using 256 facial features or points).

4. instruct the access control system to act in accordance with the identification result.

In addition to all the above capabilities, our system also determines the most fitting face recognition angle in each particular case. The system can also issue a warning if a visitor has used another person’s card to enter the premises.

There are several scenarios one should be prepared for that are almost guaranteed to be a frequent happening during the identification process.

As a dynamically growing company, we’ve noticed that you can barely ever be in time with the required adjustments to the system as new arrivals appear. Your HR folks may just be too busy with other things, and, even if your Face Recognition system is integrated with your HRM system, there is still one thing your office manager or security staff must definitely be able to perform quickly and on the fly:

• Easily add any new person to the system database.

There are several more functions that must be easily accessible to your office manager and security staff:

• They must be able to quickly adjust the system if it fails to identify a person completely (for example, due to poor photo quality) or identifies this person incorrectly (for example, by mistaking him or her for another person).

• They must be able to expeditiously re-train the system for just one single newly added person without having to wait for the whole of the system network to be re-trained overnight. This way, one will be able to prevent several similar recognition failures, occurring consecutively on the same day.

As you will have guessed, our product supports both these functions.

While working on our product, we also realized the great importance of just another system capability the system administrator must be able to easily access and use:

• The ability to quickly and effortlessly change the neural network configuration being used to one more suited to the current conditions, such as, for example, the type of source data or environment. For instance, our Face Recognition app allows quickly switching from SVM to Random Forest.

In order to be able to improve system performance, respond to a possible immediate threat, or avert a potential risk, you must be able to analyze the whole amount of the visual information your system has captured within a length of time. This can only be done if your Face Recognition application has advanced logging functionality (and this is precisely the case with our Face Recognition system).

Hence, is our next tip:

• Do not go for a Face Recognition system that doesn’t have a good logging capability.

What is a good logging capability?

In order to encompass the full variety of possible scenarios, we have implemented 3 types of logs:

1. A Movement-tracking log, - this log is created by default whenever a movement is detected.

2. An Event log, created when the system detects a face for each of the persons present on the scene.

3. A Face log, created by the system within an Event log as it identifies a face, indicating the recognition probability.

What does all this intricate logging allow and why do we think it is an absolute must that any Face Recognition app have ALL of this functionality?

Of course, we haven’t spent so much time and effort on our app’s logging capability just for the kick of it. The benefits our approach has generated are nothing short of mission-critical. Here is what this functionality allows you to do:

1. You can see what was taken out of or brought onto your premises.

2. You can see if several persons have entered using one card and determine who the cardholder is.

3. and, most importantly, you can create various alerts that will help respond to any possible risks in a more timely and expeditious manner.

As a good Face Recognition system does not only secure your premises but is also able to help you leverage the workload on your infrastructure.

• Never pick a Face Recognition system that doesn’t support advanced reporting.

To keep our clients from making this mistake, we’ve equipped them with a comprehensive set of reports.

In some industries, a Face Recognition system can be used very gainfully for purposes other than its main purpose of securing a company’s premises. That is why,

• If you are in the supermarket or food service industries, check if the Face Recognition system you are looking at includes any interesting marketing features you could put to good use.

What can those features be and how can they be used by businesses in the said industries?

Well, some of those we have already implemented at this point (Body Recognition and Tracking) allow major retailers to perform Basket Analysis and determine the number of visitors to their different departments. The latter allows them to stagger the workload on their cash register staff and achieve greater customer satisfaction.

In turn, restaurants will be able to determine their more frequent visitors and center their marketing efforts around those insights. They will also be able to identify those of their clientele, who require more attention due to the not so exemplary conduct they have shown during their previous visits.

Similarly, if you run a Web-based business of some kind that calls for stringent user identification, some of the more advanced Face Recognition systems can help. Our one would allow you to identify visitors to your website using their IDs.

Disclaimer: The present article reflects solely the subjective viewpoint and findings of the SYTOSS AI Development Team on the topics, covered herein and does not represent advice to buy or not to buy, or use or not to use any software product or technology.

The above article provides some generic information on how Face Recognition technology can be used by businesses and organizations to improve their security and operations. It lies solely in your responsibility to make sure that all the appropriate laws and regulations are observed by you when you use it for this purpose.