Most buildings in Taiwan are existing, mid-sized and working to a tight budget. They are already full of cameras — yet almost nobody is watching them. This year we came to the expo to answer one question: can these already-built properties enjoy AI without tearing anything out?
Booth theme: smart operations for existing buildings — Seeing × Understanding
Conversations about intelligent buildings usually assume new-build projects, generous budgets and a formal rollout. Walk into a real housing complex or factory office, though, and the conditions are different: the building is twenty or thirty years old, the equipment is already installed, there is no IT team, management fees have been frozen for years, and night-shift staff are both hard to find and hard to afford.
We headlined the booth wall with "Smart operations for existing buildings — Seeing × Understanding", and printed the three pain points we set out to solve right beside it:
- Hiring keeps getting harder — night-shift staff are hard to find and hard to afford.
- Dozens of cameras cannot be watched — one person on dozens of feeds will miss the critical seconds.
- Conventional AI alerts on any motion — false alarms pile up until someone switches it off.
Solution 1 · Seeing: AI Eagle Eye Cloud Patrol Team
Built for night-time security in residential complexes and commercial offices. It runs on Everspring Group's in-house AIoT wireless security hardware: sensors watch around the clock for smoke and fire, intrusion and emergency calls; the moment an alarm is triggered the system pulls live footage as corroboration, and our 24-hour control centre reads the scene in real time and responds according to standard operating procedure. The goal is to reduce night-shift posts without reducing protection, so communities can find a balance between staffing and budget.
Solution 2 · Understanding: AI Eagle Eye Video Patrol Platform
Built for sites that already have surveillance, but nobody watching it. The platform applies VLM (Vision Language Model) technology: rather than merely recognising objects, it reads what is actually happening in the frame. It runs scheduled automatic video patrols every day, lets you write detection rules in natural language, reviews frame by frame and pushes only genuine anomalies, stores annotated evidence clips for audit, exports patrol logs and reports in one click, and sends alerts straight to a phone or LINE.
Conventional video analytics
- Only answers "is there a person / vehicle / motion", and draws a box
- Cannot read what is happening across the whole frame
- Alerts on any motion; cannot separate normal from abnormal
- Adding a new category means re-labelling and re-training
VLM — Vision Language Model
- Reads the whole frame and describes the event in natural language
- Combines time, occupancy and environment into one judgement
- Distinguishes context: the same frame can be normal or abnormal
- New rules are written in plain language — no retraining
The three example photographs on the booth wall showed exactly what conventional analytics cannot do: "lights on late at night with nobody present" requires several conditions read together; "nothing may be stored in a fire escape route" requires understanding that someone has piled things up; "community pets must not be off-leash" alerts when an animal is set down unrestrained. None of these can be answered by "is there a person" alone.
Live hardware: AIoT wireless security devices
The left of the booth was a hands-on area: the wireless security control panel (network + 4G dual backhaul, backup battery, built-in siren, 20 wireless zones plus wired zones), door and window sensors, infrared sensors with pet-immunity, emergency buttons, indoor voice sirens, NFC patrol readers, and a fire-signal receiver that relays fire alarm signals back to the panel. Wireless transmission uses U-Net 5.0 with a range of up to one kilometre — no cabling and no damage to interiors, which is what makes fast deployment in an existing building possible.
An ecosystem: one booth, a complete service chain
The right of the booth presented our group resources and partner network. Group subsidiary Huachen Smart Property Management brings smart lockers and digital inspection systems in from the earliest planning stage of intelligent building management; Tong Tai Technology develops, designs and sells integrated smart security and property technology systems. Ecosystem partners included Sharp Energy (one-stop EV charging and EMS energy management for communities), Cheng-Deh Fire Protection (PADTEX EV fire blankets and roll-away storage cases) and Kerrus (Taiwan's first GMP-certified domestic AED manufacturer, with IoT cloud management).
Keynote: from seeing to understanding
During the expo, Gao Yu-Heng, Associate Vice President of WorldTrend Security's Planning & Development Division, was invited by the organiser to take the stage with the keynote "From Seeing to Understanding — sustainable smart security upgrades for existing buildings".
1. Three structural problems in conventional security
The talk opened with the state of the industry: a shrinking, ageing guard workforce — people are harder to recruit and costs rise every year, with night shifts and multi-site coverage permanently short; the limits of watching a screen — one person on dozens of feeds loses attention quickly, and "installed" is not the same as "watched"; and review after the fact rather than in the moment — most footage is only retrieved after an incident, by which time the loss has already happened and the recording is merely evidence.
"Office towers, housing complexes, construction sites, factories — cameras are everywhere. And almost none of that footage is ever watched."
Keynote section: the situation today
2. The same light on: normal, or abnormal?
This was the clearest comparison in the talk. At 3 a.m., someone is present and the light is on — somebody is working late, the light makes sense, so simply log it and do not disturb. At the same 3 a.m., nobody is present and the light is still on — that may be wasted energy or something left switched on, so it should raise a prompt.
Conventional AI struggles here because it can only look at "is anyone there" or "is the light on" separately; it cannot bind "time + occupancy + light state" into a single judgement, so it either floods you with false alarms or misses the case entirely. VLM expresses the whole situation in one rule — "if the light is on late at night with nobody present, treat it as an anomaly" — and the AI judges accordingly.
3. Turning house rules directly into detection rules
- Natural-language custom rules — "nothing may be stored in the fire escape route", "staff may not smoke in restricted areas": write it, and the AI watches for it.
- Multi-condition logic — "after closing time there should be no movement inside the store", "a missing hard hat only counts as a violation during working hours".
- Smart de-duplication for state conditions — the same illegally parked vehicle prompts once per night rather than on every patrol, and closes itself when it leaves.
- Occupancy-aware behaviour — when people are present, certain rules (such as "lights should be off") automatically stand down so normal work is not interrupted.
Common detection patterns today cover people and intrusion; vehicles; site safety (missing hard hats or reflective vests); environmental anomalies (flame, smoke, standing water, doors or roller shutters left open, abnormal lighting); site violations (smoking, burning joss paper, unrestrained pets); assault events; behaviour analysis (falls, loitering, running, climbing, gathering); and fully custom rules. Each can be further scoped by area and time window.
4. What AI takes over is the repetitive, draining, night-time work
The talk was explicit about a point that is often misread: AI does not replace guards. It takes over scheduled night rounds, screen watching, handwritten patrol logs and monthly inspection reports — patrolling every camera on schedule, reading frame by frame and pushing only anomalies, keeping automatic records with images and text, sending alerts to a phone or LINE, and exporting reports in one click — so that people can concentrate on what genuinely needs human judgement and human response.
5. Not just technology, but a security-grade operating process
Beyond the technology, the talk set out three operating designs at our control centre: an incident SOP checklist (fire includes notifying 119, intrusion includes notifying 110, with every step ticked, named and timestamped so handling quality can be audited); a daily handover briefing (compiled automatically at 07:30 covering the previous day's patrols, incidents, offline devices and today's watch items); and automatic alarm consolidation (the same state is merged into one card per day, and anything already handled stops nagging). Alerts are only taken seriously when they mean something.
For residents and owners, the same work surfaces as a reassurance calendar, real-time LINE notifications, stored evidence clips and transparent patrol records — so the value of the service is no longer invisible.
6. The most sustainable upgrade is the one that discards nothing
100%
Existing equipment reused
Cameras, recorders and cabling all stay — no e-waste, no duplicated investment
Zero works
No cabling, no disruption to interiors
Zero interference with a building in operation
On demand
AI compute runs only on schedule or on alarm
No 24-hour full-stream uploading
That is the angle we wanted to offer under this expo's sustainability theme: the barrier to going smart can shift from "replace the hardware" to "connect to the cloud." For an existing building, the greenest upgrade is often the one that throws nothing away.
"AI should not be something only large enterprises can afford. Bringing AI down to every building that has already been built — that is what WorldTrend intends to do next."
Keynote closing: our ambition
Four days on the floor: what people actually asked
The visitors who stayed longest arrived with very concrete questions. Building committee chairs asked how security quality holds up after night-shift posts are reduced. Facility managers asked whether obstructions in fire escape routes can be caught automatically. Property management firms asked whether patrol logs can become the monthly report directly. Plenty of people, standing in front of the live hardware, learned for the first time that a wireless panel can be armed without any cabling at all.
Those conversations confirmed something for us: the market is not asking for flashier terminology. It is asking for something that can start tomorrow and show results tonight.
