AI-Powered Cameras and Video Analytics: Beyond Simple Recording

Traditional CCTV systems record video and leave it to a human operator to watch feeds or review footage after an incident. The problem is obvious: no person can monitor dozens of camera streams simultaneously, and searching hours of recorded video for a specific event is tedious and slow. AI-powered video analytics transform cameras from passive recorders into proactive sensors that understand what they see and alert operators the moment something relevant happens. For Australian IT resellers, this shift represents a significant upsell opportunity — moving customers from commodity cameras to intelligent solutions that deliver genuine operational value beyond simple security.

The analytics capabilities available today include people counting, heat mapping, line crossing and intrusion detection, loitering alerts, abandoned object detection, facial recognition, licence plate recognition, and even slip-and-fall detection. These features are used across retail, logistics, healthcare, education, and government sectors. The technology has matured to the point where false-positive rates on quality cameras have dropped dramatically, making analytics practical for everyday deployments rather than just high-security niche applications.

AI video analytics can run in two primary locations: directly on the camera (on-edge) or on a dedicated server or appliance. On-edge processing uses a neural processing unit (NPU) or specialised SoC built into the camera itself. Cameras from Dahua, Hikvision, and Axis now ship with embedded AI chipsets — Dahua uses its proprietary WizSense and WizMind series, Hikvision offers AcuSense and DeepinMind, while Axis leverages its ARTPEC-8 chip with deep learning capabilities. The advantage of on-edge processing is that analytics run locally with no additional server hardware, reducing bandwidth and eliminating a single point of failure.

Server-based analytics, by contrast, pull video streams from standard cameras and process them centrally using GPU-equipped servers. Platforms such as Briefcam, Agent Vi, and Milestone XProtect with analytics plugins follow this model. Server-based solutions are vendor-agnostic — they can analyse feeds from any ONVIF-compliant camera — and they allow more complex analytics that require correlating data across multiple streams. However, they demand significant compute resources, typically requiring NVIDIA GPUs, and they add network bandwidth overhead because full-resolution streams must be sent to the server for analysis.

People counting is one of the most commercially valuable analytics features. Retail stores use it to measure foot traffic, calculate conversion rates, and optimise staffing schedules. Shopping centres use aggregate counts to justify lease rates and plan tenant mix. Modern AI-based people counting uses deep learning to distinguish humans from other moving objects — shopping trolleys, prams, and shadows — achieving accuracy rates above 95 per cent in well-lit environments. Dahua WizMind cameras and Axis cameras with the ACAP people counter application both offer reliable on-edge counting with dashboard integration.

Heat mapping builds on people counting by visualising where people spend time within a space. The camera aggregates positional data over hours or days and generates a colour-coded overlay showing high-traffic zones in red and low-traffic zones in blue. This data is invaluable for retail merchandising, museum exhibit placement, and workspace design. For resellers, combining people counting with heat mapping creates a compelling business intelligence package that extends the value proposition of a camera system well beyond traditional security.

AI-powered intrusion detection replaces simplistic motion detection with intelligent classification. Instead of triggering an alarm every time a pixel changes — which leads to constant false alarms from rain, swaying branches, or animals — AI analytics classify objects as human, vehicle, or other. Rules can then be set to trigger alerts only when a person crosses a virtual tripwire or enters a defined zone. Dahua WizSense cameras call this Smart Motion Detection (SMD), while Hikvision AcuSense cameras apply the same principle. The reduction in false alarms is dramatic — often dropping from hundreds of nuisance alerts per day to fewer than five genuine notifications.

Perimeter protection analytics go further by combining tripwires and zones with active deterrence. Many AI cameras now include built-in white-light LEDs and speakers that can flash a strobe and play a pre-recorded warning when an intrusion is detected. This active deterrence approach physically discourages intruders rather than simply recording them. Dahua TiOC (Three in One Camera) models combine full-colour imaging, AI analytics, and active deterrence in a single unit, making them a popular recommendation for Australian residential and small business installations.

Facial recognition is arguably the most powerful — and most contentious — AI analytics feature. The technology captures facial biometric data, converts it to a mathematical template, and matches it against a database of known faces. High-end cameras like the Dahua WizMind series and Hikvision DeepinMind NVRs can store databases of tens of thousands of faces and match them in under a second. Use cases range from VIP recognition in retail, to access control in corporate buildings, to watchlist alerting in public safety applications. The technology is mature enough that it works reliably in controlled lighting with cooperative subjects, though accuracy drops with extreme angles, poor lighting, or masks.

In Australia, facial recognition technology sits in a complex regulatory space. The Privacy Act 1988 classifies facial biometric data as sensitive information under Australian Privacy Principle (APP) 3, meaning collection requires either consent or a specific legal exemption. The OAIC (Office of the Australian Information Commissioner) has taken enforcement action against companies using facial recognition without adequate consent — notably the 2021 determination against Clearview AI and the 2022 investigation into retailers like Bunnings and Kmart. Resellers must advise customers that deploying facial recognition in public-facing environments without explicit, informed consent is likely to breach the Privacy Act. Signage alone is generally insufficient — the OAIC has indicated that passive collection via signs does not constitute valid consent for sensitive biometric data.

Deploying AI analytics successfully requires more than simply installing an AI-enabled camera. Camera placement is critical — analytics perform best when the camera has a clear, unobstructed view of the detection zone with subjects occupying a minimum pixel count (typically at least 80 pixels tall for human detection and 120 pixels for facial recognition). Overhead fisheye cameras are excellent for people counting but poor for facial recognition, which requires a near-frontal angle. Lighting conditions must be considered too — while many AI cameras support analytics in low light via IR, colour information significantly improves classification accuracy, making full-colour night cameras or supplementary lighting worthwhile investments.

Network and storage planning also changes with AI analytics. On-edge analytics generate metadata streams that need to be ingested by the VMS for searching and reporting — ensure the VMS supports metadata from your chosen camera vendor. Storage requirements may actually decrease because AI-triggered recording eliminates hours of empty footage, but metadata databases grow over time and need their own retention policies. Finally, invest time in commissioning — walk through every detection zone, test every rule, and verify alert delivery to monitoring stations, email, and mobile apps before handing over the system.