08
Service 08 · Eagle AI Cloud Patrol
Eagle AI Cloud Patrol Platform
Your CCTV records 24/7 — but no one watches 24/7. Manual patrols manage 3–4 rounds a night, miss things, and leave only paper sign-in sheets. AI cameras ask you to replace hardware, run new cabling, and pay for on-device compute.
WorldTrend's new Eagle AI Cloud Patrol Platform plugs into your existing NVR — no new hardware. Cloud AI runs scheduled snapshots and analysis around the clock, turning manual rounds into cloud patrols that are more frequent, more consistent, and never skipped. Every patrol leaves an auditable record with thumbnail and natural-language description.
ENT · PrimarySMB · Chain storesCOM · Large communities
CORE CAPABILITIES
- 1Plug into existing NVR — no new hardwareConnect via NVR API or scheduled upload — no rewiring of the existing CCTV system. The lowest barrier to entry of any AI patrol option, and our biggest advantage over "buy a new AI camera" competitors.
- 2Cloud-scheduled AI patrol5-second snapshot bursts (≈10–15 frames) on configurable schedules — far higher frequency than human rounds. No sick days, no skipped patrols, no shortcuts.
- 3Cloud-managed · always-onFully managed SaaS — no servers to run, no models to retrain on your side. The AI keeps upgrading and auto-updates to the latest capabilities, so the platform gets smarter the longer you use it.
- 4Built-in rule packs + custom promptsShips with common-sense anomaly rules (people in cleared areas, smoke or fire, standing water, doors or shutters open, lighting anomalies). Customers can add any condition in plain language.
- 5Auditable inspection logsEvery patrol auto-generates "thumbnail + natural-language description + timestamp" — replacing paper sign-in sheets with an evidence asset you can audit.
- 6Webhook + API + dashboardPush incidents to LINE / Email / Teams / Slack; route into WorldTrend's monitoring-center SOP, or integrate straight into your existing systems.
TYPICAL VENUES
Large property managers / HOAsChain officesFactories / warehousesOSH compliance auditsNighttime residential buildings
PRICING MODEL
A predictable fixed monthly fee (site fee + camera count) — no upfront hardware purchase, no need to hire a night-shift guard. Positioned to replace periodic human patrols, delivering significant cost savings with a budget you can plan around.
ENGAGEMENT
NVR site survey → API / FTP integration (1–3 days per site) → rule-pack setup → scheduled patrols go live → monthly audit report, scaled to site size.
Deep Dive
How Cloud Patrol works
DETECTION · What it can detect
Built-in scenario pack, ready out-of-the-box
🚷 Unauthorized presence during off-hours
Set business / restricted hours — if anyone is present after that, alert. Includes standing, sitting, loitering.
🔥 Smoke / fire / visible flooding
Common early-stage incident signs in plants, equipment rooms, and basements.
🚪 Door / shutter unexpectedly opened
Overnight warehouses, office access points, residential back doors left open.
💡 Lighting / equipment anomaly
Should be on but isn't, should be off but isn't — plus custom equipment-state descriptions.
🦺 OSH compliance audit
PPE (hard hats, hi-vis vests), fall-protection harness at height, hot-work violations.
📝 Any custom condition — described in plain English
Describe what you want monitored in one sentence and the platform runs it — no engineers retraining models.
VLM TECHNOLOGY · Why it's stronger than ordinary AI
Powered by VLM (Vision Language Model) — looks at images and understands you in plain English
❌ Traditional AI image analysis
- • Only recognizes pre-trained object classes (people, cars, faces, plates)
- • Rules are hard-coded by engineers (line-crossing, dwell time N seconds)
- • Want to monitor a new scenario? Retrain the model or add new hardware
- • Once it false-alarms, it keeps false-alarming — the hardware can't be corrected
✅ Our VLM (Vision Language Model)
- • Model understands scene meaning directly from images — no fixed object list
- • You issue instructions in one plain-English sentence: "Is anyone loitering in the restricted area?" "Anyone not wearing a hard hat?"
- • Adding a new monitoring rule means rewriting one sentence — no retraining
- • Click "this was a false alarm" once, and the system gets more accurate over time
FEEDBACK LOOP · False-alarm feedback loop
Gets more accurate over time — not a build-once-and-stop system
STEP 1
One-click feedback on the dashboard
Get an inaccurate alert? Click "false alarm" or "missed event."
STEP 2
Site-specific sample accumulation
The system collects false-alarm and missed-event patterns specific to your site.
STEP 3
Prompts automatically fine-tuned
Detection descriptions are optimized for your site — accuracy improves month over month.
※ Compare with hardware-based AI cameras: the model is locked in at the factory. On-site false alarms keep being false alarms — to get smarter, you have to buy new cameras.
COMPARISON · Side-by-side
Three options side by side — where do we win?
| Dimension | Eagle AI Cloud Patrol | AI camera brand | Human night guard |
|---|---|---|---|
| Upfront investment | No hardware purchase, plugs into existing NVR | NT$30,000–100,000 per camera — adds up fast across many cameras | No equipment, but rosters must be set up |
| Patrol frequency | Scheduled patrols — can be far more frequent than human rounds and consistent | Continuous real-time detection | 3–4 rounds per night shift; fatigue easily leads to missed rounds |
| Cost structure | Predictable fixed monthly fee, plannable | Large upfront hardware purchase + ongoing maintenance | Highest monthly labour cost |
| Accuracy evolution | Gets more accurate over time — false-alarm feedback auto-tunes | Locked in at factory; false alarms remain false alarms | Depends on the person's experience |
| Audit record | Auto-generated every patrol (thumbnail + description + timestamp) | Only when an event is triggered | Manual paper sign-in |
| Capability boundary | Strong at absolute anomalies; weak at split-second motion | Hardware-defined | Depends on staff |
※ Honest positioning: this platform replaces "periodic manual patrols," not 24/7 continuous watch. Scheduled snapshots can miss split-second events between two captures — the same constraint as human rounds, complementary to continuous monitoring.
USE CASES
What customers actually use it for
- 1Nighttime residential building patrols · replace the guard's 3–4 nightly rounds with auto-generated "thumbnail + description + timestamp" audit logs.
- 2Warehouse cleared-area monitoring · alert on any person present after-hours; concurrently monitor smoke/fire, flooding, and doors/shutters left open.
- 3Multi-site chain office oversight · one dashboard for every branch; events push into LINE groups via Webhook.
- 4OSH compliance audits · hot work, working-at-height, PPE-compliance rule checks for plant operations, with evidence retained for every patrol.
FAQ
Frequently Asked
Don't see your question? Get in touch and we'll answer based on your specific environment.
Which NVRs / cameras does the platform support?
Technically we are brand-agnostic — any NVR or IP camera with standard RTSP streaming and a fixed IP (or remote-reachable via public-IP + port forwarding, VPN, or vendor P2P) can be integrated. Most mainstream NVRs on the market meet the requirements, and no equipment replacement is needed. Every site is surveyed before integration to confirm the connection method and camera placement.
How is this different from "Eagle Eye Video Surveillance" and "AI Eagle Eye Cloud Patrol"?
Eagle Eye Video Surveillance: alarm-triggered video verification with monitoring-center dispatch (event-driven). AI Eagle Eye Cloud Patrol: residential-community service combining multi-sensor coverage, central dispatch and patrol vehicles (community-only). This platform: cloud SaaS that plugs into your existing NVR, runs proactive scheduled patrols, and produces auditable logs (replaces manual rounds). The three are complementary, not overlapping.
What events is the AI good at — and what is it not good at?
Strong at absolute / common-sense anomalies: people in cleared areas, smoke or fire, standing water, doors or shutters open, lighting anomalies. Can describe behavior at the level of standing/sitting/lying, walking, loitering, direction of entry or exit, opening doors and moving objects. Weak at split-second motion (instant of a fall, a punch, an object thrown across frame) and at fine-grained gesture recognition. Positioned as "description + coarse classification" — sufficient for patrol-style anomaly judgment, complementary to continuous live monitoring.
What about personal data and video privacy?
Footage is pulled only during scheduled windows and deleted after analysis (we do not retain raw video long-term). Event records keep only the thumbnail and text description. Cloud workloads run on GCP asia-east1 (Taiwan), with a data-processing agreement signed with each customer.
What is VLM? How is it different from ordinary AI image analysis?
VLM (Vision Language Model) is a model that looks at an image, understands the meaning of the scene, and describes it in natural language.
Traditional AI image analysis only recognizes pre-trained object classes (people, cars, faces, plates), with rules hard-coded by engineers (line-crossing, dwell time N seconds). To monitor a new scenario, you have to retrain the model or add new hardware.
This platform uses VLM, so customers issue instructions in one plain-English sentence: "Is anyone loitering in the restricted area?" "Anyone not wearing a hard hat?" "Has the shutter been opened?" Adding a new rule means rewriting one sentence — no model retraining, no camera swap.
Traditional AI image analysis only recognizes pre-trained object classes (people, cars, faces, plates), with rules hard-coded by engineers (line-crossing, dwell time N seconds). To monitor a new scenario, you have to retrain the model or add new hardware.
This platform uses VLM, so customers issue instructions in one plain-English sentence: "Is anyone loitering in the restricted area?" "Anyone not wearing a hard hat?" "Has the shutter been opened?" Adding a new rule means rewriting one sentence — no model retraining, no camera swap.
Will it produce false alarms? How are false alarms handled?
Yes. Every AI system produces false alarms in the early days; our difference is that false alarms feed back into the system — you can flag "this was a false alarm / missed event" with one click on the dashboard. The system collects samples specific to your site, then automatically fine-tunes the detection prompts for that site, and accuracy improves month over month — it gets more accurate over time.
Compare with AI cameras: hardware-based AI is locked in at the factory. On-site false alarms keep being false alarms — to get more accurate, you have to buy new cameras.
Compare with AI cameras: hardware-based AI is locked in at the factory. On-site false alarms keep being false alarms — to get more accurate, you have to buy new cameras.
What does it cost? How does it compare to hiring a night guard or buying AI cameras?
A predictable fixed monthly fee, quoted case by case after a site survey (depends on site, camera count, rule complexity).
Compared with hiring a night-shift guard: one night-shift headcount costs about NT$50,000–70,000/month — this platform's monthly fee per site is typically far below that.
Compared with AI cameras: vendors ask you to pay NT$30,000–100,000 per camera upfront (which adds up to hundreds of thousands across many cameras). This platform has no upfront hardware investment, a predictable monthly fee, and rules you can rewrite at any time.
Compared with hiring a night-shift guard: one night-shift headcount costs about NT$50,000–70,000/month — this platform's monthly fee per site is typically far below that.
Compared with AI cameras: vendors ask you to pay NT$30,000–100,000 per camera upfront (which adds up to hundreds of thousands across many cameras). This platform has no upfront hardware investment, a predictable monthly fee, and rules you can rewrite at any time.
Next Step
Want to see how Eagle AI Cloud Patrol plugs into your existing NVR?
Book a 30-minute consultation. We'll review your sites, budget, and staffing model and recommend a fit — or use our service matcher to self-serve a starting point.