Artificial Intelligence (AI) has become an incredibly hot topic recently, thanks to public facing tools like ChatGPT. AI has the potential to revolutionize many aspects of our modern-day world, and security solutions are no different. AI could fundamentally change how we look at security, as we observe its applications in facial recognition, anomaly detection, and predictive analytics. Security solutions supported by AI are boasting faster response times and more accurate threat detection, which makes them valuable for several industries.
As the number of cameras being used to monitor both public and private property grows world-wide, there is an increased demand for the use of artificial intelligence (or “AI”) to leverage the metadata in video streams, that can be utilized when using intelligent IP video products. These IoT systems create massive amounts of data every second of every day, for a wide range of reasons. For surveillance devices, this data can be used for security, convenience, or emergency situations. Whether they are gathering data for a business to improve its operations or receiving alerts due to a security or safety issue, cameras are evolving into proactive business and security tools instead of simply providing forensic or reactive investigation capabilities.
Much of this surveillance data has potential value, but it first must be transferred, processed, stored and analyzed. The most common current model sees all data transferred from the connected device to a data centre or to a server for storage and analysis. Since not all data will be useful or valuable, its transfer and storage can create significant wasted resources in bandwidth and memory, not to mention the upstream energy consumption and cost for housing centralized processing power. Enter edge computing.
Edge computing puts greater processing power at the “edge” of the network or, in more tangible terms, within the network video camera itself. This allows for a level of data analytics by the device and therefore the transfer of only meaningful, useful data, or data which requires further analysis (for example, alerting officials of exceptions at border control where passport verification is required). The benefits in bandwidth and storage requirements are obvious, let alone those in increased efficiency in operations. As data transfer often necessitates compression, edge computing can circumvent this, and, when combined with AI-based analytics, can provide the clearest possible image for AI to work from.
Importance of the Edge
The “Edge” of a network refers to computing infrastructure that is closer to the sources of data within a given system. In an AI-based security solution, this is usually the video surveillance devices themselves, which have over the years become increasingly powerful and capable network computing devices using high quality lenses as their primary sensors to collect data. Edge computing can be an important aspect of AI-based security solutions, as it reduces latency through proximity and helps keep sensitive data local, which can reduce the risk of data breaches. For example, edge solutions may function without a server or cloud-based connection, meaning that a single malicious actor cannot gain access to an entire system by compromising one aspect of its operation.
The additional benefit of running analytics at the edge, especially when it comes to cameras and analytics, is the ability to run analytics on an uncompressed image, resulting in higher accuracy and metadata. In traditional server-based deployments, the camera compresses the video stream first to conserve bandwidth, and the server-based analytics are performed on the compressed stream.
Enhancing Security Solutions with AI
There are several ways in which AI can be deployed to increase the security of a given facility. Facial recognition, anomaly detection, and predictive analytics all contribute to a more secure environment which can react faster to threats. At the highest level, AI-based security solutions detect threats faster and with greater accuracy; this has numerous real-world applications such as detecting fraudulent financial transactions or identifying illicit items in security x-rays.
Facial recognition technology, for example, can be used to identify individuals who are entering a secure area, or to assess a group of people and identify individuals who may pose a safety risk. In the United States, the Transportation Security Administration (TSA) has tested facial recognition powered by AI, to scan the faces of passengers and compare them to a database of known safety threats. Lower-level solutions come in the form of something like “Face Detector” software meant to deter thieves by giving the illusion that they are being tracked in a retail space through the sharing of an audio message that informs passers-by that they are being watched.
Of course, we cannot talk about AI and facial recognition without addressing privacy and restrictions on collecting personally identifiable information (PII). Options include solutions for static privacy masking, which is ideal for indoor or outdoor scenes with fixed areas that aren’t allowed to be monitored. Then there’s dynamic masking that uses an edge-based privacy shield application on visual cameras that allow users to see movements or activities while safeguarding privacy in real time.
Another type of AI-based security solution is anomaly detection. This utilizes AI to detect patterns of behaviour and identify behaviour that is outside of a learned norm. This is generally beneficial in combatting users who are accessing data or are in areas they shouldn’t be, perhaps for malicious purposes. A user who constantly attempts to enter a secure area they are not authorized to use could be worth looking into, for example.
Predictive analytics are another aspect of AI-based security solutions. By identifying patterns, AI models can predict events that may pose a security risk. In the case of financial fraud, patterns of money laundering or other types of scams can be analyzed through past examples, and then these same patterns can be detected early on, potentially saving people from falling victim to fraud or scams.
AI analytics at the edge
For surveillance systems utilizing edge analytics, it is possible for cameras to detect that something or someone is moving within a given scene. This footage could then be analyzed by a human actor to understand exactly what the entity is, and if they present a security risk. However, incorporating AI analytics at the edge and training models to detect and classify various entities within a monitored area can have incredible safety and security benefits.
Running analytics on a server on location has provided us with performance advantages. Now, powerful on-board processing offers new solution benefits on the edge. Edge analytics are video analytics that process and analyze video data right on the camera, close to where it’s captured rather than on a server or in the cloud. This kind of on-board deep learning capabilities allow solution providers to offer unique opportunities like developing third-party AI based applications that can solve problems. Some of these applications can track people and send alerts if someone is in a location for longer than a specified time and does not require that the people be active or moving to be detected. And all this data runs entirely on the camera, with no additional server. It uses true machine-learning to identify people and the period they were there. It basically tracks and sends alerts. This demonstrates that server-based solutions are not the only option when there’s AI or other high-performing analytics as part of a security solution. Surveillance cameras that are fitted with Deep Learning Processing Unit (DLPU) chips, where analytics are installed directly on the camera, are considered more today due to their simplicity, scalability, flexibility and reduced cost.
Incorporating AI analytics at the edge can greatly reduce the rate of false positives, and as a result reduce the need for human intervention and more efficiently route those resources to situations which need timely and appropriate responses. As an example, surveillance cameras on motorways, backed by AI analytics at the edge, could clearly identify objects or accidents and alert drivers. These cameras could differentiate between vehicles and people, and accurately alert both drivers and emergency services to situations unfolding in real time.
As AI becomes increasingly common and more advanced, so do threats to cybersecurity. AI-based security solutions will eventually become a necessity in the modern world as they work to keep individuals and organizations safe from threats. More research and advancement are needed to ensure that AI-based security solutions have the public’s best interests at heart and can be used responsibly.
In terms of edge computing, scalability and accuracy will only increase in years to come. We have already seen massive leaps in terms of capability and processing power, so it is exciting to imagine the advances in object detection and analysis that are sure to emerge soon.