Join or Sign in

Register for your free 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

Facing challenges in face recognition: one-to-one vs. one-to-many

Facing challenges in face recognition: one-to-one vs. one-to-many
Facial recognition is often part of our daily lives whether we realize it or not. For facial recognition technology to work there are certain conditions that must be met.
Facial recognition is often part of our daily lives whether we realize it or not. For facial recognition technology to work there are certain conditions that must be met. These conditions play a critical role in the success of how well facial recognition works, particularly in a one-to-many facial recognition system where the system is matching faces against a database of photos instead of just one photo. Unfortunately, oftentimes the conditions needed for optimal performance are the technology’s biggest challenge.

“Facial recognition can work very effectively in public spaces, but there is a need to change the positioning and focal length of some of the video surveillance cameras to get better views of the faces,” said Brian Lovell, CTO of Imagus Technology.

“One-to-one facial recognition systems are easier to use and do not require extensive logistics and staffing,” said Roger Rodriguez, Manager of Image Analytics at Vigilant Solutions. Not only are they easier to use, the environment in which they are used makes it easier to identify faces. “One-to-one face recognition systems can be set up in a controlled environment, with sufficient lighting, where the user interacts with the camera and assumes an optimal position,” said Elke Oberg, Marking Manager of Cognitec Systems.

In addition to easier set up, Rodriguez pointed out that one-to-one systems tend to be more security oriented. “Examples are cell phone or computer access via facial recognition,” he said. “Upon initial set up, a user is required to submit a still photo into the database for facial recognition comparison. Since the dataset is one single picture of the subject, a one-to-one system only looks for that one face. The accuracy rates are always higher because the template matching algorithms are comparing only on face to one image.”

On the other hand, “One-to-many systems need to be very carefully designed and installed to capture quality faces without the knowledge of the persons of interest,” said Lovell. “Simply retrofitting existing video surveillance systems with face recognition with invariably produce poor results.”

Challenges for face recognition in real-time surveillance
Oberg identified the three major challenges of face recognition technology for real-time surveillance applications in public spaces as: non-cooperative users, low/changing lighting conditions, and existing equipment/infrastructure. “Face recognition performance still drops to significantly lower accuracy levels in low light or with extreme light changes (bright sunlight); when the face is turned or tilted more than 20 degrees, and when the image quality is below 60 pixels between the eyes. Users often want to use existing cameras and networks, leading to lower grade images and/or slow response times,” she added.

The importance of camera placement used for facial recognition, especially in larger crowd situations, was also highlighted by Rodriguez. “Camera placement and image capture quality are critical components to the success of facial recognition systems. If images originate from lower quality video surveillance camera feeds and are also off axis, the still image will most likely not be ideal for facial recognition,” he said. “If cameras are set at a distance to capture large crowds, this is not conducive for facial recognition because subject faces would be too far during image capture.”

In order to overcome some of these challenges, Rodriguez said, “If camera systems are set to capture faces at face level and have pan, tilt, and zoom capabilities, then these frontal position captures meet the criteria for searching.” He explained that normalized poses of face capture are conducive for searching, leading to higher accuracy rates for obtaining possible matches in facial recognition search results.

Companies have also improved their own technologies to overcome low-quality images and difficult lighting. Vigilant Solutions’ facial recognition technology corrects the deficiencies found on many types of uncontrolled images with tools that focus on the enhancement of pose, lighting, and occlusions. This allows the end user to correct problematic images that would normally be rejected by other systems. Cognitec’s video screening product includes a camera control feature for certain camera models that improves the camera’s grain and exposure time for the best possible illumination of a face found in the video stream.

Imagus Technology has taken on these challenges by reducing the need for expensive cameras. “Imagus allows the cost-effective deployment and interlinking of multiple face recognition systems in public spaces. This means that there is a much better chance of recognizing persons of interest because they can’t hide from all the cameras all the time,” Lovell said.

Another way companies are facing these challenges is by working with end users to help them choose the optimal camera models and locations for the best result.

Improvements in technology
There have been many improvements in one-to-many facial recognition technology in the last few years; improvements that have made detecting faces in crowds more reliable.

“All face recognition vendors are working on improving the algorithm performance for video material, in particular for lower resolution images and partially hidden/covered faces,” Oberg said.

One way companies are doing this is through cooperation with public safety agencies. “Through routine collaboration, public safety agencies work hand in hand with the biometric developers of the software and have collectively developed better searching algorithms, which have significantly reduced false positive rates,” Rodriguez said. “These collaborative sessions also formalize industry standard best practices, and establish public safety facial recognition guidelines for agencies to follow.”

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

Share to: