New AI Secures Public Spaces Using Security Cameras, Drone Surveillance, and Facial RecognitionNovember 09, 2019 by Gary Elinoff
The ever-advancing field of artificial intelligence is opening new vistas for securing homes and public spaces alike.
In the past, we've discussed how AI has become more prevalent in public safety and security domains—for instance, in predictive policing. In this article, we will assess how AI is changing the nature of theft detection, aerial security, and facial recognition. To illustrate how AI is shifting the direction of security in each of these areas, we will narrow in on a few programs and applications that lean heavily on AI-based technology.
SK Telecom's AI Inference Accelerator to Prevent Break-Ins
SK Telecom's AI inference accelerator (AIX), implemented on Xilinx Alveo accelerator cards, can detect home or commercial break-ins using deep neural networks. AIX enables SK Telecom's detection service, T view, to monitor thousands of customers' commercial and home camera systems in real-time. The deep neural networks allow the AI accelerator to process massives amounts of data from the cameras; once T view detects physical intrusion, it can dispatch security guards or other law enforcement to the scene.
Xilinx's Alveo U250 data center accelerator card. Image from Xilinx
Xilinx’s Alveo cards achieve low detection latency and achieve greater throughput (frames per second) than possible with GPUs. As a result, SK Telecom can sharpen T view's security capabilities even more—exactly what is called for in a modern data center. The Alveo cards fit into a standard PCIe slot.
Deep Sentinel Monitors Porches, Stops Burglaries
There are several applications that combine the efforts of AI and human security personnel, as is the case of Deep Sentinel. The company sets what it describes as an “intelligent perimeter” of cameras around a private home. The images from the cameras are streamed to a central location equipped with AI that can determine if a threat exists.
Next, a live agent at the central location comes online. If, for example, a burglar attempts to steal an Amazon package, the agent issues a verbal warning. If that doesn’t do the trick, the agent calls the local police.
Deep Sentinel combines AI capabilities with the efforts of human surveillance and law enforcement. Image from Deep Sentinel
Without artificial intelligence, there is no economical way for live agents to watch more than a few homes at a time. With AI separating hundreds of irrelevant events from the potentially criminal events within moments, the purview of a single watcher is exponentially expanded.
Robotics and Drones
Airobotics’ Automated Industrial Drones Provide Real-time Aerial Surveillance
Before AI-based solutions became available, companies wishing to take advantage of the power of drones were faced with limited options. Using a third-party company to operate a drone is unreliable, and training local technicians is expensive.
Airobotics created a solution to monitor and secure large industrial facilities with virtually no human intervention. Using automated industrial drones, decision-makers can collect aerial data and access visibility to surveil a facility.
Automated drones employ AI to monitor industrial facilities for biohazards and physical intrusion. Image from Airobotics
The solution includes the drone, its airbase, and the company’s AI-based software. Drones are launched and landed automatically with no human pilots needed. The software integrates distinct data types, which are automatically translated into clear security insights in real-time.
AI-Empowered Drones to Guard Europe’s Frontiers
The Roboborder project, expected to come online in 2020, involves swarms of drones patrolling the EU’s borders. The goal is to send unmanned mobile robots to border and coastal areas, where they will utilize optical, infrared, and thermal cameras and sensors to detect criminal threats. The main idea is to have as many sensors in the field as possible to assist border patrol personnel.
The Roboborder sensor robots while use AI to distinguish threats. Image from Roboborder
One of the more controversial features of this endeavor will be the use of the drones—equipped with artificial intelligence—to decide what constitutes a threat before any human intervention.
Evolve Edge Screens Large Crowds for Security Threats
Unlike a typical metal detector and bag check system, the smart checkpoint system can detect in real-time where a threat is on a person's body, and, because of AI, the device gets "smarter" over time. To date, it has scanned more than one million people worldwide.
This technology can also match the faces of individuals as they walk by, which prevents a human traffic jam around a stopped checkpoint. Operators can further assess non-matched, unrecognized individuals while allowing recognized visitors, such as employees and known individuals to simply walk through.
Evolve Express screens a number of people at once, discerning unrecognized visitors from recognized employees. Image from Evolv Technology
Evolve Express is made possible by the company’s Evolv Cortex AI Software Platform. The technology can screen 60 people per minute and allows groups through multi-lane entrances that can be up to 8-feet wide.
FacePro Identifies Hard-to-See Faces
FacePro is Panasonic’s facial recognition system that can identify difficult-to-recognize individuals in surveillance footage: that may include individuals who wear a hat or glasses, individuals who appear aged compared to an identification photo, or people whose faces are angled away from the camera.
Most of the currently available systems require that the camera must view the individual straight on, but FacePro can read faces at angles of up to 30° up or down, or even up to 45° to the left or right.
FacePro is designed to detect individuals despite indirect angles, facial obstruction, or age. Image from Panasonic
This AI-driven technology allows law enforcement to identify and track suspicious individuals or persons of interest, who move through locations monitored by FacePro. For instance, specific faces, such as known shoplifters, can be registered.
Police agencies can also provide Facepro users with pictures of wanted criminals. Alarms can then be installed to notify human security guards when FacePro identifies a suspect's face.
Facepro also features an analytics section, which can be invaluable in conducting market research. Users of the software can count the number of individuals present facility-wide or in a given sub-area and note how long they remain in the area. It also allows operators to detect age and gender.
With machine learning becoming an integral part of theft detection, aerial security, and facial recognition, it seems that engineers will see an increasing demand for AI capabilities.
What are your thoughts on the intersection of AI and physical security? Share your thoughts in the comments below.