Join or Sign in

Register for your free asmag.com membership or if you are already a member,
sign in using your preferred method below.

To check your latest product inquiries, manage newsletter preference, update personal / company profile, or download member-exclusive reports, log in to your account now!
Login asmag.comMember Registration
https://www.asmag.com/showpost/34324.aspx
INSIGHTS

Applying deep learning to facial recognition cameras

Applying deep learning to facial recognition cameras
Face recognition deep learning is more effective and accurate than classical algorithms in terms of face recognition, and deep learning algorithms is evolving, too.
More and more, AI and deep learning algorithms are applied to facial recognition to make it more effective and accurate, especially in less-than-ideal situations. Similarly, with advances in camera hardware, deep learning is embedded in facial recognition cameras also.
 
Deep learning entails the computer extracting features by itself, with little to no manual intervention. The more features it extracts, including features that are hard to describe, the more accurate the recognition process becomes. That’s why facial recognition engines are increasingly employing deep learning to improve accuracy.
 
“Over the past five years, AI technologies based on neural networks have almost completely overshadowed everything else. Face recognition has become much more reliable, especially in unfavorable conditions,” said Alan Ataev, Global Sales Director at AxxonSoft. “At the time being, AI and deep learning underlie all of the most effective solutions on the market. These technologies beat classical algorithms in terms of recognition quality. And if you look at the speed at which they are evolving, no one would seriously consider applying classical algorithms.”
 
“The core task in face recognition is to take an image of a face and convert it into a set of features. You want the features generated from two images of the same person to be as close as possible (regardless of lighting, expression and other distractions) while making sure that two images of different people produce significantly different features. Given enough data and computation, a neural net can simply do a better job of this than a hand-designed system. The network can end up using features that are more complex and counterintuitive than a human designer is likely to consider,” said Michael Jamieson, Computer Vision Scientist at Senstar. “This change in technology allows newcomers to be very competitive in the face recognition market since accumulated institutional knowledge of previous techniques is less important.”
 
And increasingly, vendors of facial recognition cameras have also put deep learning into their products. These include Hikvision’s DeepinView Dual-Lens Face Recognition Camera and Dahua’s 2MP Starlight Face Recognition Box Network Camera, both embedded with deep learning functions.
 
“Some of the most direct benefits that deep learning algorithms can bring include achieving comparable or even better-than-human pattern recognition accuracy, strong anti-interference capabilities and the ability to classify and recognize thousands of features,” said Jerry Hong, IP Product Manager at Hikvision. “With deep learning technology, the average accuracy of face recognition increases significantly – by 38 percent. Our advantages in artificial intelligence in security have come about by having our own excellent algorithm development team and using the most powerful GPUs in our computer platforms.”
 
“Artificial intelligence algorithm is mainly to complete the face recognition and comparison of the entire process. Deep learning algorithm, based upon the training of big data, can improve the accuracy of face recognition, which then can be applied in more complex environments, such as poorer image quality and wider angles,” said Stanley Hu, Product Director at Dahua Technology.
 

Tips on buying, installing face recognition cameras

 
When selecting facial recognition cameras, several things need to be looked at to get the right product at the right price. “System integrators should pay more attention to the recognition accuracy, the frontend face gallery capacity, the maximum number of faces under a single screen, installation requirements (height, angle, recognition range) and adaptability to complex project environment,” Hu said.
 
Meanwhile, how to set up the cameras to get the optimal results is also important. “System integrators should deploy the cameras based upon actual parameters of the camera, the type of camera and the actual business needs of customers. Usually they need to avoid poor environment such as strong backlight, large bevel angle, darkness, distance, unstable base installation and unsteady power supply,” Hu said. “Priority deployment should be in major roads facing the stream of people. For example, Dahua smart capture system is suitable to be installed in the entrance to park and shopping mall entrances, with a suggested installation height of about five meters, which delivers multi-target face capture within 50 meters.”


Product Adopted:
Surveillance Cameras
Subscribe to Newsletter
Stay updated with the latest trends and technologies in physical security

Share to: