Creating a Safe and Secure Society with
High-Precision Facial Recognition Systems
Using the world’s highest level of facial recognition accuracy developed using deep
learning technology to make security monitoring and management more efficient
As use of the Internet of Things (IoT) and artificial intelligence (AI) rapidly expands, a wide range of information is beginning to be linked together in diverse industries to make our lifestyles more convenient. At Panasonic, particularly in the field of security, deep learning technology using AI is integrated with intelligent camera technology and video monitoring software to advance the technical development and implementation of facial recognition systems.
The World’s Highest Level of Facial Recognition Accuracy
Facial recognition systems can be effectively utilized in the field of security monitoring and management. Facial recognition can be applied to a variety of solutions: from identifying terrorists, suspicious persons, shoplifters and dangerous people, to expediting access to secure facilities. Panasonic has been engaged in the technical development of security camera systems for more than 60 years, constantly staying on top of the needs of those at the front lines of the field. Panasonic began R&D of image sensing technology close to 30 years ago, developed the LUMIX brand of digital cameras, and also started sales of facial recognition systems paired with our security cameras in 2011, thus accumulating a considerable track record and expertise in the field.
Recently, as interest in the field of security has increased together with rapid changes in social conditions, in August 2018 Panasonic launched its high-precision facial recognition system “FacePRO” with deep learning technology, which combines security and image sensing technologies. The recognition engine of this system was developed together with the National University of Singapore — one of Asia’s leading academic institutes, which conducts cutting-edge research on deep learning technology. The system has been certified as the world’s highest level of facial recognition performance*1 by the National Institute of Standards and Technology (NIST), a leading standards institute of the United States.
*1 Having achieved the world's highest level of facial recognition performance, as certified by the National Institute of Standards and Technology (NIST) of the United States (as of April 2017) in a comparison test (IJB-A Face Verification Challenge Performance Report, IJB -A Face Identification Challenge Performance Report).
Even with “1:n” Verification, High Recognition Accuracy Is Achieved
There are two types of facial recognition: “1:1” (one to one) verification using one’s face to match against a photo ID such as a passport; and “1:n” (one to many) verification which matches a target individual’s face from within a number of unspecified people. The Panasonic facial recognition system uses a unique algorithm that combines a machine learning method based on deep learning and a similarity calculation method that suppresses errors. Even in the event of “1:n” verification, the facial features are extracted to greatly improve the recognition performance.*2 Highly precise verification can identify faces that are difficult to recognize with conventional technologies, including faces at an angle of up to 45 degrees to the left or right or 30 degrees up or down, and those partially hidden by a surgical mask or sunglasses. High-precision verification is also achieved with aging changes (recognizing faces that have changed over the years and with different hair styles).
*2 Improves facial recognition performance by up to 500% compared to conventional systems. (Rejection rate reduced to 20% when the wrong person acceptance rate is set to 0.01 with the IJB-A face image dataset.)
From Passive to Active Use: Preventing Incidents and Accidents
Compared with conventional security systems, which are often just used to confirm events after they have occurred — such as to use recorded images as evidence — this system distinguishes the facial image of persons who require attention in advance. It can also be used to determine whether those persons are still present by conducting searches in real time. This allows users to actively prevent incidents and accidents, or leads to the quick discovery of such, all while helping to make security monitoring more efficient and increase safety and security.
Reducing the Server and Network Loads
With ordinary facial recognition systems, all of the captured images are sent to a server, and facial detection and recognition occurs within the server, which places all the data processing burden on the server. In addition, because a wide bandwidth is required to send the high volume of images, and a large-capacity hard disk drive (HDD) is needed to store them, the system typically tends to be large-scale.
With FacePRO, the camera crops only the images (“Best Shot”) that are necessary for facial recognition from all the facial images recorded of the same person passing in front of a camera, and sends only those best shots to the server. As the camera bears part of the large image analysis processing load typically done by the server, and because only the images that are best suited to facial recognition are sent, there is less data transmission required by the network. This reduces the server load to approximately 1/5th that of a conventional system, and reduces the network load to approximately 1/10th. Compared to a conventional system that does not use this function, a 40-50% reduction in the overall system costs can be achieved.
Integrated Management with an Existing System
With Panasonic’s “i-PRO Monitoring System” for instance, because integrated operation is possible with the existing system, the need to operate and manage a completely separate facial recognition system is eliminated, and the work can all be centralized in an integrated system. For example, when doing a target person search, which is an important part of a crime prevention monitoring system, by searching for a designated person, information on when and where that person appeared, which direction they were headed, and where the person presently is, can all be shown on a single monitoring screen in a timeline. The facial detection, facial verification, and tracking of captured images from the recorder are all done on one graphical user interface (GUI), enabling one-stop integrated management. This ability to provide each necessary component, from cameras to entire systems, is a strength offered only by Panasonic.
Facial recognition systems not only streamline security monitoring and management work, but they can also be used for marketing applications. Panasonic participates in demonstration projects in a variety of fields and deploys products and systems for both consumer and enterprise use. Moving forward, we will continue to offer cutting-edge image technologies and integrated management as easy-to-use solutions to meet our customers’ needs.