What to ask before you choose a facial recognition solution

Facial recognition offers several applications that can help customers. But knowing which brand or solution to choose may be difficult for many. Quite a few companies offer facial recognition solutions now, and many more may add it to their portfolio in the coming days.
 
With algorithms becoming more complex and the hardware improving every day, remaining up to date on the developments in this sector is difficult for customers. So when they need to purchase a solution, their criteria are limited to factors like the size of the brand and access to its sales team.
 
But asking the right questions is essential to ensure that you select a solution that meets your needs in the long term. Speaking to asmag.com recently, Dean Nicolls, Chief Marketing Officer of Oosto, listed seven questions customers should ask before buying a facial recognition solution.

1. How accurate is your solution?  How long does it take to recognize an authorized employee (or someone on a watchlist)?

 No company will tell you that their facial recognition is less accurate compared to the competition. It is up to you to check the rate of false alarms and detection speed they offer to know the quality of the solution.
 
"Real-time facial recognition is a challenging technical problem, and not all solutions are created equal," Nicolls said. "Oosto offers the highest level of facial recognition accuracy with 0.1 percent false alarms and 0.2ms detection speed."

2. How well does facial recognition perform in less-than-ideal environments (e.g., when someone is not looking directly at the camera)? 

 When you check the accuracy levels that the vendor provides, it is essential to remember that they are probably recorded in ideal conditions. Unfortunately, most customers can't offer these ideal conditions on their premises, and hence it is important to ask how the platform works in the wild.
 
"Most facial recognition systems do not perform well in the wild when the conditions are suboptimal, that's why testing these solutions with just a few cameras can quickly separate the contenders from the pretenders," Nicolls said.

3. Can I use my existing camera infrastructure? 

Many customers already have a security infrastructure in place, and they just want to add facial recognition now because the technology has become better and more useful. In such a scenario, you need to ask the vendor if their solution can work on the cameras that you use.
 
"Ideally, you can deploy facial recognition software that uses computing power efficiently and requires minimal changes to your visual sensor infrastructure," Nicolls pointed out. "This requires the software to integrate with leading camera manufacturers (e.g., Honeywell, Johnson Controls)."

4. Can I track people across multiple locations?

Many businesses and organizations have large campuses that they must protect. They would need facial recognition solutions that can be managed from a central location, allowing the operators to monitor people throughout the premises.
 
"For instance, many hospitals are spread out and encompass many buildings," Nicolls said, speaking from the point of view of customers in health care. "Ideally, hospital administrators can centrally manage authorized medical staff and watchlists across locations and control how POI (person-of-interest) data is managed, analyzed, and distributed while receiving real-time alerts." 

5. How does your solution protect people's privacy?

Ensuring that the facial recognition solution does not violate anyone's right to privacy is important. Many companies offer features like bystander blurring (in a hospital setting), dynamic data retention times, and hard data deletes.  

6. Does your solution perform equally well across different demographics, or does it suffer from ethnic bias?

Recent studies have raised concerns about some facial recognition algorithms being biased to people of certain demography. While this is an issue that most major companies have addressed, customers would be better off asking about it before purchasing.
 
"Demographic bias is largely overstated in the press," Nicolls said. "Better facial recognition solutions train their AI models on large, representative, balanced data sets. Neural networks - built with diverse skin tones, facial poses, genders, ages, and ethnicities - perform better in the wild and minimize demographic bias."

7. How can we lower our total cost of ownership (TCO)? 

It's important to understand all the costs required to deploy a facial recognition system, including the software, hardware (servers), power, cooling, and IT management.  Modern facial recognition solutions are starting to leverage edge computing to push more analytic capabilities closer to where data is collected, helping organizations achieve greater responsiveness and efficiency.
 
This enables organizations to deploy facial recognition video processing on near-edge compute appliances, effectively shifting the GPU-heavy compute load from expensive on-premise servers to small, dedicated power-efficient devices.