From common CCTV cameras to autonomous security drones, visual monitoring devices are everywhere. These security systems continuously produce high volumes of footage, much of which sits unused once it's been captured. It's nearly impossible for humans to monitor multiple live security feeds and take proactive action.
This is where AI comes in - computer vision technology leverages the abundance of visual data to identify what data is useful, what can be ignored, and what demands immediate attention.
In this article, we at SiaSearch have put together the most promising AI applications for security, as well as the most innovative computer vision startups within each.
Unlike human personnel, computer vision-based security systems are able to watch security footage tirelessly, monitor everyone in view, identify patterns and any suspicious activity. In recent years, a large number of startups have stepped up to provide AI-powered video surveillance.
Umbo is a Taiwan and San Francisco-based startup that provides cloud-based video security systems for businesses. Umbo's smart security cameras, in combination with proprietary computer vision-based software, autonomously detect and identify suspicious events, such as intrusion, tailgating, and wall-scaling.
Deep Sentinel has built a similar solution for home security. Whenever a threat is detected, Deep Sentinel streams live video to real human security personnel to remotely intervene via microphone.
In response to the uptick in mass shootings events, Actuate (formerly known as Aegis AI) integrates with existing camera feeds to automatically identify anyone who's brandishing a firearm. Once the model identifies a weapon, it alerts security teams and law enforcement.
"We can detect a weapon before a trigger is pulled," said Ben Ziomek, Actuate co-founder and CTO. "In some instances we can enable a security response before any bullets are fired."
A number of computer vision startups offer innovative solutions to restrict or allow access to certain areas or facilities.
San Francisco-based Swiftlane, for instance, uses facial recognition to allow or deny access to offices, apartment complexes, and other physical spaces. The solution employs deep learning and computer vision techniques to provide a single-sign-on using a mobile phone or video intercom. After signing up, users only need to look at the face reading terminal to unlock the doors to areas they are authorized to enter.
Similarly, Paravision's platform is designed to be used by global security device manufacturers, solution providers, systems integrators, and financial services firms in situations where an error could have profound negative consequences.
Computer vision technology has also made great strides in detecting and identifying threats at security screening checkpoints.
For example, Silicon Valley's Synapse Technology automates security screening, enabling checkpoints worldwide to catch more threats while reducing operating costs and increasing throughput. Their platform, Syntech ONE®, integrates with new and existing checkpoint machines at airports, courthouses, federal buildings, and more.
Evolv Technology's Edge system uses a combination of camera, facial recognition and millimetre-wave technologies to scan people walking through portable security gates at airports. Machine learning models automatically check for threats, including explosives and firearms, while ignoring non-dangerous items.
Inspecting vehicles can be a challenge for checkpoint security teams.
Difficult to access, the undercarriage is an ideal spot for adversaries to hide illicit materials such as explosives, weapons, and drugs. UVeye is an Israeli startup that provides an automated under vehicle inspection scanner that captures high-resolution images as it scans passing vehicles. Using advanced deep learning algorithms, the scanner is able to detect and flag anomalies in seconds.
Computer vision can also help retailers react to theft and threats as they happen. With roots in MIT's artificial intelligence labs, StopLift analyzes security video and POS data to distinguish between legitimate and fraudulent behavior at checkout. By applying advanced computer vision algorithms to existing camera feeds, StopLift's ScanItAll system tracks items that pass through the checkout lane, associate them with POS, and flag any suspicious activity.
Developed by Japanese telecommunications company NTT East and startup Earth Eyes, AI Guardman is an automated security camera designed to catch shoplifters in the act. Based on open-source technology developed at Carnegie Mellon University, AI Guardman scans live video streams from cameras in convenience stores and supermarkets, tracking every customer inside. When a threat is detected, the system sends an alert to shop staff in real time.
Advances in computer vision technology can also be used to address common public safety concerns, from cutting crime rates to slowing down the spread of infectious diseases.
Atlanta-based startup Flock Safety aims to reduce crime in America by 25% by using computer vision to improve public safety in neighborhoods. Their automated license plate reader (ALPR) software, FlockOS, combines character recognition with computer vision and machine learning to provide real-time insights to crime prevention authorities in over 1200 US cities. It uses patented Vehicle Fingerprint™ technology that identifies a vehicle even if it's modified or if its license plate is missing or covered.
Furthermore, in the wake of the COVID-19 pandemic, many public facilities have equipped existing security cameras with AI-based software that can track compliance with health guidelines.
Reducing the risk of infection in stores is a major priority for brick-and-mortar retailers in particular. Retail analytics platform Aura Vision provides a suite of COVID-focused solutions to promote in-store safety, from features that monitor face mask compliance to heat maps to visualize high traffic areas since the last clean.
Not only has computer vision proven its practical value for physical security on private property and public spaces, it has also demonstrated value on a national scale, with applications ranging from environmental monitoring to use in military systems.
Shield AI is a company that works with federal, state, and local departments and agencies to deliver next generation surveillance systems. Their first product, Nova, is a Hivemind-powered drone that searches buildings while simultaneously streaming video and generating maps.
Finally, Orbital Insight specializes in applying computer vision to geo-analytics. The company uses satellites, drones, balloons, UAV footage, and geolocation data from mobile phones to analyze human activities and provide businesses and governments with key behavioral insights and address security concerns.
In the security sector, the challenge is not data acquisition, but effective data management. The majority of security camera recordings aren't useful for the relevant business or computer vision functions, causing extremely high redundancy and low information density.
SiaSearch makes it easy to select subsets of highly quality training data, helping you build better ML models at lower costs.
With SiaSearch's lightweight API, users can:
- Automatically index, structure, and evaluate raw data captured by security cameras and sensors
- Visualize data and analyze model performance
- Search and access all security footage across all events and attributes
- Identify rare edge cases and curate training datasets
Interested in learning more? Reach out to SiaSearch for a free proof of concept.