The question at the heart of the report is a practical one. AI video tenders on large public infrastructure routinely run from tens of millions to hundreds of millions of NT dollars, yet more than half of all housing in Taiwan is over 30 years old. How are those existing buildings supposed to afford the same technology?

Why this subject

The report begins with two pressures arriving at the same time. The first is the labour structure of the security industry: the minimum wage keeps rising, the workforce is ageing, turnover stays high and young people are reluctant to join, while residential management fees have been frozen for years. Night posts are becoming both unaffordable and unfillable.

The second is the intelligence gap in existing buildings. Citing Ministry of the Interior figures, the report notes that of roughly 9.22 million dwellings in Taiwan about 4.83 million are more than 30 years old - over half - while fewer than a quarter are under 20 years old. More tellingly, many projects that once earned green building or intelligent building certification lost their smart systems after the developer handed over and withdrew: the property managers and security firms that followed neither understood nor could maintain them, so within a few years the systems fell out of use and the building reverted to conventional management.

The three barriers the report identifies

01
Costs out of proportion

Spending on safety earns the client nothing extra, and its benefit is hard to show in the accounts. Budgets for AI security on public infrastructure reach tens to hundreds of millions of NT dollars - a scale with no feasibility at all for a shop, a small business or an apartment building.

02
Patrols that cannot be verified

Conventional night patrols are logged with RFID checkpoint readers, but the tags can be taken down, scanned all at once and put back before the shift ends. The log looks complete while no real patrol may have happened all night - the client pays and has no way to verify.

03
Nobody dares touch the wiring

In a building of twenty or thirty years the wiring is tangled, and clients fear that touching one thing breaks another. If adopting AI requires replacing every camera with an AI model at NT$30,000 to NT$100,000 each, a single site easily passes a million - and most clients drop the idea there.

WorldTrend's answer: remove the barriers rather than raise the spec

The report sets out four design decisions, each made so the solution can stand up inside an existing building.

1. Connect to the existing NVR; keep the cameras

The platform connects directly to the client's existing network video recorder; standard RTSP streaming plus a fixed IP or remote access is enough. For the few purely analogue sites with no network port at all, the approach is to leave the original system untouched and add a digital recorder alongside it that can output a network signal. Footage still records on the client's own equipment while one stream is split off for analysis - existing hardware stays in service, and no rewiring can break it.

2. Use a VLM, so rules can be written in plain language

Earlier AI video analytics recognised only pre-trained objects, with rules hard-coded by engineers; the model inside a hardware AI camera is fixed at the factory, so once it produces a false alarm on site it keeps producing it. With a vision-language model (VLM) the system understands what is happening in the frame, and the user can state the condition to watch for in one plain sentence. Adding a rule needs no retraining and no new camera.

The report gives three real rules to show what this means in practice:

3. Patrol, rather than monitor continuously

This is what brings the cost down. The system captures automatically on a schedule during set windows, at a frequency far above the three or four rounds a person walks in a night, with no absences, no shortcuts and no missed points. Compared with continuous 24-hour analysis, scheduled sampling needs far less compute. Every round automatically produces a record with a thumbnail, a natural-language description and a timestamp, replacing the paper sign-in sheet with auditable evidence.

If something happens between scheduled rounds - a smoke, door, window or flood sensor triggering - the system inserts an extra patrol immediately. The operator in the monitoring centre no longer sees a single line of text reading "client X, smoke alarm" but live footage already interpreted by AI. Confirmed fire means dialling 119 directly; a reading of someone smoking means alerting on-site staff to check, with the response graded to the event.

4. A monthly fee, with no upfront hardware purchase

The monthly fee is calculated from the number of cameras on site, the patrol frequency and the complexity of the rules, with no upfront hardware spend. Clients set the window and frequency per camera - more rounds on critical points, night-only on others - and control the cost themselves.

A solution only has room to spread if it works without replacing cameras or altering existing wiring.
The feature report, on the barrier to adoption

Results: how much is saved, and how clearly

30–50%
Flexible half-post: no staff on site at night, guards kept by day for parcels and admin, the monitoring centre watching remotely after dark
20–30%
Reduced night post: night staffing cut from two shifts to one, with a dedicated monitoring platform on site backing up the monitoring centre
Auditable
Every round leaves a thumbnail, a description and a timestamp, turning unverifiable night service into a traceable record

The report also gives the basis for comparison. Against a monthly cost of roughly NT$50,000 to NT$70,000 for one night-shift guard, the platform fee for a single site is usually well below that. On upfront cost, because the platform connects to the existing NVR instead of replacing hardware, the barrier falls sharply compared with buying AI cameras at NT$30,000 to NT$100,000 each, several of which can add up to hundreds of thousands. The monthly fee is fixed and predictable, and the rules can be changed at any time.

Two extensions now under way

Catching up on operations in already-certified buildings

The solution is already in service at the Tongchuang Enterprise Building in Taipei. The building earned both green building and intelligent building certification years ago, but measured against today's standards its level of intelligence now falls short. The report suggests that using AI video technology to close the operational gap in already-certified buildings offers a useful reference for certification renewal and for policy on upgrading existing buildings.

Flame detection linked to residential EV chargers

WorldTrend also works with EV charging solutions provider Sharp Energy, fitting sensors beneath residential chargers that detect a specific flame wavelength. The signal enters WorldTrend's system and reaches the monitoring centre; once confirmed by phone or video, power to that charger or the whole charging circuit can be cut automatically before notification proceeds. The report notes that this can compress the time from incident to the arrival of firefighters from as long as 20 minutes to under 10.

WorldTrend's track record, as set out in the report

The report further observes that where parent company Everspring Industry mainly serves the intelligence of new construction, WorldTrend Security concentrates on adding intelligence to existing buildings - a position that gives it a close understanding of the constraints and needs of sites already in use.