AI and edge computing

A key concept for AI edge computing is placing the “brains” (AI running on computers and devices) close to the “eyes” (cameras). Adding “ears” (audio sensors) means that AI trained to recognize patterns involved in security incidents can perform that pattern matching at a speed and scale that humans can’t — starting instantly on-site. The pattern-matching work can be distributed across cameras, video servers, purpose-built local appliances, and powerful cloud-computing platforms.

This article examines different types of AI-enabled security solutions, focusing on the major and minor roles that AI edge computing plays in each. It does not attempt to describe in detail the features or functions of each solution. Instead, the goal is to explain each solution’s distinct approach and how it incorporates or interacts with AI edge computing. An exception is the Alcatraz section, which gives added attention to privacy regulations and the handling of personally identifiable information (PII) — topics that remain poorly understood within much of the physical security profession and the industry that serves it. It is also worth noting that all the companies featured have patents pending or awarded related to their AI-enabled devices and platforms.

Surveillance and monitoring historical shortcomings

One of the core challenges with earlier generations of in-house and third-party alarm monitoring has been the prevalence of two major weaknesses: false positives (alerts that are not security incidents) and false negatives (no alerts for actual security incidents). The historically high rate of false positives is the root cause of the common operator fatigue in surveillance and monitoring operation centers.

Smarter and faster security response

Most of the solutions below focus on massively reducing false positives and achieving virtually no missed threats. They support diverse monitoring and response approaches, covering today’s varied surveillance needs. Large organizations may apply multiple solutions at some sites to address all security risks.  

All the solutions mentioned below can be applied to existing security system deployments. The use cases explored include:

  • Large-scale third-party of central alarm monitoring operations
  • AI edge readers for multi-sensor access control authentication 
  • Enterprise-grade full-site real-time situational awareness
  • Video intelligence AI for scenario-specific needs
  • Small and medium-sized business protection during regular and off-hours

Actuate: Going beyond object classification for extreme accuracy and response

Central station alarm monitoring, serving small businesses and homes for over 150 years, supports over 13,000 U.S. stations with intrusion detection, fire monitoring, video surveillance, access control, and environmental monitoring. Historical shortcomings push remote monitoring toward proactive solutions.

Actuate’s cloud-based AI video analytics transforms monitoring centers into high-tech command centers and remote guarding into AI-augmented services. It detects intruders, weapons (99% accuracy), fires (earlier than sensors), and critical crowd formations, integrating with most cameras, network video recorders/video management systems (NVR/VMS), and monitoring platforms.

Measurably better false alarm reduction

Typical object classification analytics (detecting motion or people/vehicles) prioritize avoiding missed threats, causing high false positives. Actuate’s AI models and scenario-focused AI training excel in context-aware processing, reducing false positives by 95%+ while maintaining virtually no missed threats, backed by a $10,000 reimbursement policy for missed detections.

This reliability allows staff to manage more sites with less stress, with one center reporting a 57% per-site alert reduction (over 400,000 fewer monthly).

Uniquely for a monitoring platform, Actuate accommodates both live video streaming from cameras and (NVR/VMS) setups, as well as workflows where edge AI in cameras or recorders sends images via email — using the simple mail transfer protocol (SMTP) — based on scene or object motion detection. Supporting both streaming and edge AI email alerts is crucial because some sites cannot stream video due to bandwidth limitations, and some have workflows built around email alerts. Actuate’s AI models apply the same stringent processing to video images, clips, and continuous streams, achieving uniform accuracy across all monitoring station customers.

Actuate doesn’t depend on AI edge computing. However, edge AI in cameras and NVR/VMS systems does in fact act as an initial filter for alerts sent to Actuate. Thus, AI-enabled analytics in such cameras and NVR/VMS systems be tuned for zero false negatives—since Actuate does the work of assuring the lowest possible false positive rates for monitoring center processing.

Alcatraz: Privacy-first edge AI for frictionless access control

Alcatraz leverages advanced edge computing through its Rock product line, offering frictionless facial authentication for secure facility access. Here, “frictionless” means users can simply approach an entry — without stopping to present a badge, enter a PIN, use a phone or provide a fingerprint or iris scan. Rock and Rock X devices incorporate AI-powered 3D sensing and mathematical facial modeling, authenticating users passively and seamlessly, so no physical interaction is necessary. Additionally, Rock devices alert on tailgating events — which is a valuable real-time capability that is unusual for a reader-type device.

Core access control functions

Before exploring Alcatraz’s technology, it’s important to understand the two foundational steps in physical access control and how they relate to biometric authentication and privacy regulations:

  1. Authentication: Verifies the identity of the user.
  2. Authorization: Determines what the authenticated user is permitted to do.

Step 2, Authorization, is performed by the access controller connected to whatever authentication devices or systems are being used.

There are many methods to perform step 1, Authentication, for user identification. Examples include:

  • Keypad PIN entry
  • Card or fob electronic scan
  • Facial scan
  • Fingerprint, palm, iris, or retina scan
  • Sensor-based pedestrian gait scan
  • Voice analysis

During the access control enrollment process, the images, scans, or recordings generated by biometric methods are stored in a system database for future use in user identification. In the U.S., definitions from the Department of Defense and the National Institute of Standards and Technology (NIST) classify the scan data collected by these technologies as personally identifiable information (PII).

This classification applies because biometric data — whether stored in the access control system or a dedicated biometric system — is part of the user’s access control record, which uniquely identifies the individual. Additionally, biometric scan results (such as facial images or voice recordings) can be viewed or played back to visually or audibly confirm a person’s identity.

How the Alcatraz technology is privacy compliant

Alcatraz doesn’t store facial images and explains that their mathematical facial template is not humanly identifiable PII under strict privacy definitions, on the basis that:

  • It cannot, by itself, reveal or be used to reconstruct the facial likeness of an individual.
  • It is only usable within the specific context of their access control product, in conjunction with enrolled badge IDs, and is never used for broad surveillance or open identification.
  • The template is not usable out of context — an external party with access to the encrypted mathematical model alone would be unable to determine who it represents.
  • Additionally, Alcatraz never shares or transfers any PII with other systems. This means Alcatraz never has access to information like names, birthdates, and similar data—and conversely, other systems (such as the access control system) never receive Alcatraz Profiles (i.e., templates). All PII is segregated between Alcatraz and other systems, and is, of course, encrypted both at rest (AES-256) and in transit (TLS 1.2).

Thus, Alcatraz’s solution — including the Rock and Rock X — has been approved for use in several U.S. jurisdictions where image-based facial recognition is banned. The key distinctions are:

  • Explicit consent and enrollment: Users actively enroll by providing their information; no covert or open-ended data collection occurs.
  • Purpose limitation: The mathematical model is used strictly for authentication (one-to-one or one-to-few comparison), not for mass identification, surveillance or law enforcement tracking.
  • Regulatory testing: Alcatraz’s systems have been independently tested and certified to comply with national and international privacy standards, including BIPA (Illinois), CCPA (California) and GDPR (EU).

Technically accurate terms would be “facial image authentication” and “facial mathematical-model authentication” to distinguish typical facial recognition technology from the Alcatraz technology. However, without knowing the technical details (above) about the Alcatraz solution, those terms are meaningless on their own. Thus, it’s easier to do what Alcatraz does, and call their own approach “facial authentication” and the rest “facial recognition” because the second term is already in popular use.

All authentication and AI-based analytics occur locally on the Rock device, eliminating the need to transmit any facial images to the Alcatraz cloud. This is a pure application of cloud-managed AI edge computing. The Rock product line integrates with existing access control systems via standard protocols (Wiegand, OSDP), enabling organizations to add facial authentication strategically without costly equipment replacements or additions. The Alcatraz cloud platform adds further value through secure remote user enrollment, automated firmware updates, and tools that enhance privacy, efficiency and compliance.

Ambient.ai: Enterprise caliber real-time risk identification

The Ambient.ai cloud platform is engineered for early threat detection on premises and to provide actionable alerts to an on-site security team. It is purpose-built for large, high-security and compliance-intensive environments. Its core approach is to watch all cameras all the time, applying a growing library of more than 150 threat signatures — part of their patented and continually evolving intelligence model, not just a static ruleset.

This adaptive approach allows Ambient to (a) achieve zero false negatives, (b) eliminate 90% or more of false positives, and (c) alert on actual risk situations at the earliest point where recognition is possible. The platform is designed for full enterprise integration — enabling seamless coordination with GSOC workflows, access control systems, alarm infrastructure and incident response protocols.

Ambient.ai delivers contextualized real-time alerts that describe the activity or situation, enabling law enforcement and internal responders to respond with high situational awareness, supported by incident evidence and access to relevant live video feeds. Ambient’s architecture also supports compliance with data governance, privacy regulations and audit requirements common in sectors like healthcare, finance and critical infrastructure.

To do this cost-effectively and in a timely manner, Ambient uses a high-performance on-premises server (i.e., AI edge computing) to apply its threat signatures to video streams. When a match occurs, compact information about what was detected and when is sent to the cloud platform. The cloud then handles additional AI processing, analytics, and event correlation — linking related events from different cameras and sources to reveal patterns that may not be obvious in isolation.

In a physical security system, an “event” could be a door held open too long, tailgating, unusual activity detected by a camera, a badge scan where the presenting individual doesn’t enter the protected area but someone else does, or an intrusion alarm. Event correlation links these based on factors like time and location to determine whether they’re part of the same incident or threat — or part of a pattern of security violations. Event correlation over time is a sophisticated capability — and one of many unique functions that define Ambient’s advanced video intelligence platform.

The cloud platform tightly integrates with access control, alarm systems, GSOC workflows and automation frameworks to enable coordinated real-time response — such as automatically locking down areas, activating lighting or initiating two-way audio communication with trespassers — quickly placing now-situationally-aware human responders in the best position.

Ambient’s use of edge computing enables large-scale, high-performance AI processing for deployments with high camera counts and activity levels, where timely, well-informed response is critical.