From Night Patrol to Energy Inspection
Making the CCTV Already Installed in Ageing Buildings
the Front Line of Smart Energy Saving

Building smart energy management into a new development is straightforward. The hard part is the vast majority of Taiwan's building stock — ageing buildings, where retrofitting intelligence after the fact is daunting in both engineering and budget terms. And yet those are exactly the buildings that need energy savings most. WorldTrend Security's starting point is disarmingly simple: these buildings already have a pair of eyes.

Why this topic: what stands in the way of smart energy saving in ageing buildings

A brand-new building can plan an entire smart monitoring and energy automation package during construction — naturally enough. Retrofitting an older building, by contrast, usually means running new cabling, breaking into the structure, and replacing equipment, with cost and downtime as the barriers.

In his talk, Gao Yu-Heng made clear that WorldTrend is not claiming it can turn an ageing building into a fully intelligent one overnight. Instead, he offered a different angle: these buildings have long relied on property managers and security officers walking several rounds of every floor each night — checking whether lights are off, doors are closed and air conditioners are switched off, and looking out for anything unusual along the way. The eyes (the existing surveillance system) and the habit of patrolling are both still there.

The ceiling of sensors: they know "something happened" but not "what it is"

WorldTrend's foundation is the wireless IoT smart security system developed in-house by Everspring and certified to EN 50131-G2 — no cabling, no damage to the building structure, deployed at key points such as entrances, equipment rooms and vehicle ramps. But the talk was equally candid about three limitations of sensors:

• Point coverage — detection happens only where a sensor is installed; everywhere else is a blind spot.
• They report triggers, not context — a door opened, something moved; but who it was and whether it actually matters still requires video and human review.
• They cannot see "waste" — they catch intrusion, flooding and fire, but cannot read energy situations such as "a light left on in an empty room" or "what temperature the air conditioner is set to".

Extending from "security events" to "energy inspection" requires AI that understands the whole scene — and that is precisely the layer video plus a VLM adds.

Using a VLM to let AI "read" the picture

The difference between a VLM (vision-language model) and conventional image recognition was the section the talk spent the most time on:

Three worked examples: the point is not "on or off", it is "is anyone there"

Example 1: late at night — are the lights off?

Each camera is analysed on a nightly schedule at a fixed time (say 23:00). Lights already off is normal, so nobody is disturbed; lights on with a person in the frame may simply be someone working late, and is also treated as normal. Only lights on with nobody in the frame is flagged as an anomaly, prompting a notification for the manager to go up and deal with it.

Example 2: what about the air conditioning — and at what temperature?

Older air conditioners are not necessarily smart or network-connected, but a camera can "read" the control panel. A unit that is switched off, or switched on with someone using the space, is normal; air conditioning running with nobody in the frame calls for someone to go and switch it off. The system can even read the set point shown on the panel: if it falls below the required value (for example 20°C where the policy is 26°C), that too is listed as an anomaly.

Example 3: doors left open, and whether anyone is still around

A single round covers every entrance and common area, checking for security risks and energy waste at the same time. An exterior or common-area door left open for a long period with nobody around is both a loss of heated or cooled air and a security gap; lights and air conditioning still running after closing time with nobody present is a full night's electricity bill.

How it is delivered: connect to existing surveillance, live in four steps

The whole approach rests on no construction work and no hardware replacement: it connects to the existing recorder rather than installing new cameras; it captures scheduled frames from each camera every night rather than analysing continuously around the clock; a two-stage flow — a fast gating pass followed by detailed interpretation — saves compute and improves accuracy; and it leaves an auditable record, notifying the manager via LINE or the app only when there is a genuine anomaly.

What it means for decarbonisation: energy savings without a major retrofit

• No new hardware — upgrading through the existing surveillance system avoids the embodied carbon and electronic waste of manufacturing and installing new equipment.
• Waste made visible — lights left on in empty rooms, air conditioners running for nobody, and set points below policy all become actionable notifications.
• Pragmatic on compute — scheduled snapshots instead of continuous 24-hour analysis, combined with two-stage interpretation, keeps inference energy low, so the AI itself is efficient too.

The talk closed with a real example from WorldTrend's work at a 20-household community in Taipei: originally staffed with guards around the clock, the community faced rising security costs and a squeezed management fee budget. Switching to daytime guards plus overnight Eagle Eye AI — integrating the existing cameras and adding emergency call buttons — means alarms are AI-triaged and relayed to the monitoring center in real time, with a direct call to 110 / 119 where needed. Safety was not compromised, and costs were halved.

Forum agenda

This page was prepared by WorldTrend Security based on the content of the talk given on the day. Agenda details are subject to the organiser's official announcements.