To realize a safe, secure, efficient society using IT, it is important to accurately grasp phenomenon that occur in the real world. In particular, technology that acts as a substitute for human vision by analyzing still and video images to understand the real world will be increasingly important in the future. NEC Central Research Laboratories is engaging in R&D of the world's most advanced image recognition and real world sensing technologies, such as face recognition, human behavior analysis and object recognition, using the foundations of image processing, pattern recognition and machine learning technologies that we have fostered over many years.
World-beating face recognition technology
Face recognition technology is a technology that is used to identify a person from images. This processing implements high-level analysis to detect "where a face is" within an image, find "where the characteristic features are in the person's face," such as the eyes, nose, and mouth, before finally determining "who the person is" from the detected features.
NEC's face recognition technology won first place for the past three consecutive evaluations conducted by NIST for still image face matching in 2009, 2010, and 2013. NIST has just placed NEC No.1 once again in the most recent evaluation, this time for video face recognition of non-cooperative subjects (FIVE: Face in Video Evaluation) published in March 2017. This result proves the huge advantage of NEC's technology in various practical circumstances.
Examples of face recognition technology applications
NEC face recognition technology has been implemented in the border control systems of over 40 countries around the world and combines strict identity verification with improved speed and efficiency. We have also provided a face recognition entrance gate system to Universal Studios Japan® to enable "Face pass entry"—a smooth entry that makes everyone feel like a VIP. This combination of security, safety and efficiency with a VIP vibe is a trademark of NEC's solutions in this field.
Human behavior analysis (crowd behavior analysis & people flow extraction)
Human behavior analysis technology executes analysis and tracking by detecting and extracting only areas of images and video showing footage of the target person. Anomalous activity discovered in surveillance video footage can be immediately reported and the flow of people within stores and plants can be analyzed to improve sales promotions and work efficiency. We are also working on video retrieval technology that incorporates this technology and are developing a technology to enable stable analysis to be conducted even in environments with variable lighting.
"Crowd behavior analysis" to detect anomalies in crowded environments from changes in the movement of entire crowds
NEC's original technology, "crowd behavior analysis" grasps crowding conditions and detects anomalies not in specific individuals but from video images of crowds taken by security cameras in public spaces and large-scale facilities. This technology focuses on the impact of the occurrence of an incident or accident, or signs pointing to the occurrence of one, on group behavior and the surrounding crowd within video images of people, and accurately grasps and analyzes "changes in the movement of the entire crowd" linked to an anomaly. Please see this Feature Article for more details on this technology.
"People flow extraction" to grasp with high accuracy the movement of people using pre-installed cameras
The "people flow extraction" technology developed by NEC in November 2014 can grasp with high accuracy the movement of people using pre-installed cameras within stores or plants. This technology enables understanding of flows with high accuracy by linking footage taken using multiple pre-installed cameras and analyzing and obtaining flow data to grasp flows without identifying individuals by using features such as the coloring of clothing. This can lead to improvements in work procedures within plants and warehouses, and enable high level purchasing behavior analysis (marketing) of customers in stores. Please see the press release for more details on this technology.
Object recognition technology (image recognition and Fingerprint of Things)
Object recognition is technology that recognizes what a specific object is by extracting only the target area from a complex background image. NEC Central Laboratories has realized world-beating, high-speed, high accuracy object recognition using our original machine learning method, the quantifying general learning vector ("GLVQ") method.
High-accuracy, real-time image recognition using mobile phone terminals
NEC has developed high-accuracy, real-time image recognition technology that is able to determine what an object is simply by pointing a mobile phone camera at it, without using lettering or barcodes. This technology can be applied to a diverse range of target objects, from the flat surface of magazines, posters or product packaging, to natural objects and three-dimensional items. This technology has already been commercialized as the "GAZIRU®" image recognition service, a system enabling access to websites simply by pointing your smartphone at a magazine page.
The world's first "Fingerprint of Things" technology used to identify individual industrial products or parts
In November 2014, we successfully adapted our image recognition technology to develop "Fingerprint of Things," a technology used to identify individual products instantly and accurately by recognizing the "fingerprint" of individual product items that are difficult to discriminate between with the human eye. Fingerprint of Things uses the internal camera of a smartphone or tablet terminal to capture an image of an object and check it against pre-registered Fingerprint of Things image data. Because this technology can be used to achieve advanced management of individual parts, it is expected to be applied to a wide array of fields, from distribution traceability and authenticity judgments to parts management and maintenance work management. Please see the press release for more details.
mIDoT that enables ink dots to become identification tags
NEC has developed the "micro-sized Identifier Dot on Things" (mIDoT; pronounced "my dot") technology that enables single dots written with off-the-shelf glitter ink pens to become distinctive identification tags.
mIDoT can be easily applied to a wide range of objects by automated machines, or even by hand, enabling the technology to be conveniently used by anyone, anywhere to identify products or property.
This technology is expected to be used for a broad range of applications, including identification tags for ultra-small electronic components and products which are too small for the use of barcodes; tags for managing goods that are lent or taken out; keys and tickets used for access control; and identification tags for linking physical objects with digital data in the cloud.
See our press release for the details.