Facial recognition is a type of biometric technology that enables operators to rapidly identify people that appear in a video. With facial recognition software, operators can extract facial features from faces in live or recorded video and compare them against a database or watchlist of digital images (or extracted facial features) of persons of interest. If a face in a video matches one on the watchlist, a real-time alert can be triggered automatically to notify security staff who can assess the match and determine how to respond. Whether they are VIPs warranting more personal engagement or suspects who require closer oversight, persons of interest can be proactively monitored and approached or apprehended as is appropriate. In addition, a “whitelist” of authorized persons can be leveraged for triggering alerts when people who aren’t on that list are detected.
Beyond detecting potential threats or validating authorized individuals, face matching technology can be used for locating missing persons; identifying and factoring out employees when people counting in a retail scenario; and accelerating post-event investigations.
Traditionally, those who operate facial recognition solutions are law enforcement or physical security professionals, who work for either municipalities, government agencies or private corporations, such as banks, casinos, or retail stores. As the video surveillance industry has evolved with ultra-high definition cameras, combined with the growing sophistication of artificial intelligence applications, facial recognition technology has become more available, robust, and accurate – and thus more widely adopted.
Two Types of Facial Recognition
The marketplace for facial recognition is evolving and booming worldwide, and consumers have their choice of technology to implement. Of course, not all facial recognition technologies are equal and not every solution is appropriate for all settings and purposes. Face recognition technologies can be divided into two solution types: those for cooperative access control scenarios and those for non-cooperative, “in the wild” surveillance scenarios.
Access control refers to those implementations that primarily serve to manage entry of certain people to certain areas of a facility. For example, to protect the assets of a bank vault, visitors may be required to pass through a facial recognition terminal in order to verify their identity. “In the wild” face recognition encompasses face matching using CCTV video cameras intended for monitoring an area. This type of face recognition is more challenging because subjects are not necessarily looking directly at a camera and cameras are not necessarily optimally positioned or provide high enough resolution to ensure a high level of accuracy.
In addition to these two scenarios of facial acquisition, there are two ways of matching the acquired face with reference images. In controlled settings, a person usually supplies the system with an identifying object (such as an ID Card) which tells the system against which face reference the person must be matched. This type of recognition is called 1:1 matching or “verification,” because the acquired face is being matched to a single predetermined reference image. If the system has no prior identifying data about the person to be recognized (which is usually the situation in the case of “In the wild” facial recognition), the system must try to match the face against an entire watchlist (or a large subset of it). This is called 1:N matching (with N being the size of the watchlist that is under comparison) or “identification.”
When comparing stated accuracies of facial recognition engines, it is critical to check whether these accuracies were attained in 1:1 or 1:N comparisons, and in controlled or “in the wild” scenarios.
Face recognition is extremely valuable for identifying persons of interest in video; however, it is not always possible to use facial recognition technology. First and foremost, some municipalities and countries do not permit the use of face recognition technology due to privacy regulations. Yet, even where face recognition can be used, video surveillance cameras often capture faces “in the wild,” in less-than-ideal conditions: In some cases, a camera installed isn’t optimally placed or doesn’t record high enough resolution video to enable face recognition. In other cases, subjects may be facing away from a camera, have partially obscured faces, or they may be walking quickly in low-light conditions, all of which compromise the ability to provide accurate face recognition. Without optimal conditions the technology cannot always deliver the desired results.
When choosing a facial recognition solution, organizations should consider how the technology will be used, to ensure that it will meet their unique needs. Buyers should ask themselves whether their current infrastructure is suitable to enable face recognition, and whether additional or replacement hardware is needed to support the basic image quality requirements. They should also consider whether existing integrations will need to be updated or reconfigured to support the new face recognition capabilities.
Advantages of Comprehensive Video Content Analytics
In cases where facial recognition technology cannot be operated effectively – and especially where its use is illegal – organization’s efforts to locate or identify persons will be stymied, and investigations will be stalled. In these jurisdictions, organizations would benefit from a comprehensive video content analytics solution that is powered by artificial intelligence and Deep Learning, which offers other video search and alerting capabilities in addition to facial recognition. Based on Deep Learning techniques, advanced video content analysis software can detect, identify, extract and catalog objects in video footage according to classes and attributes such as gender, appearance similarity, color, size, and direction of movement. This functionality enables operators to search for objects or people using filters such as “men with black hair and a red short-sleeved shirt walking South”.
Comprehensive video investigation software also enables users to configure real-time alerts based on class, attribute, or facial combinations, so that security or law enforcement staff can be notified when someone in a video camera feed matches a description. Another benefit is that, over time, aggregated video metadata can also be visualized by a video content analytics system – presenting long-term data within dashboards, heatmaps, and reports to empower managers to make better data-driven decisions for long-term planning.
By integrating a comprehensive and extensible video content analysis solution that includes face recognition, organizations can enjoy the benefits of both solutions, applying the technology as permitted and useful to transform video into data that is searchable, actionable and quantifiable.