Camera Management System

Laravel + Filament control plane for workplace cameras, occupancy, attendance, and AI-assisted alerts.

This application does not decode RTSP video or run face recognition inside PHP. Cameras stream to an NVR (Frigate / go2rtc). That layer detects people, then this app stores events, matches faces, flags anomalies, evaluates custom vision prompts, and reports from aggregated daily data.

[Cameras] --RTSP--> [Frigate / go2rtc]
                         |
                   snapshot + webhook
                         v
              [Laravel + Filament]
                    |        |
              queues      MySQL / Redis
                    |
         Face service · Vision API · Attendance devices

Default integrations are mocks, so you can demo the full pipeline with zero cameras.

Admin branding: Aether Cam — Filament light and dark mode (user-menu sun/moon). AI via imrandevbd/laravel-ai-hub.

Demo roles (password: password)

Email	Role	Sees
super@cms.test	Super Admin	Everything + /ai-hub Studio
admin@cms.test	Admin	Full control plane + Studio
ops@cms.test	Operations	Registry, cameras, copilot, alerts
security@cms.test	Security	Monitoring, AI triage
hr@cms.test	HR	People, attendance, briefings
viewer@cms.test	Viewer	Read-only command center

AI automation (Laravel AI Hub)

Works without API keys (demo answers). Add Gemini/OpenAI/Claude keys in /ai-hub (Studio has its own light/dark theme) for live calls with failover and cost tracking.

• AI Copilot — natural-language questions over live occupancy, cameras, alerts, attendance
• Daily / shift / executive briefings — scheduled at 07:00 or on demand
• Alert explain — auto-triage on new anomalies + manual “AI explain”
• Camera vision prompt author — drafts a per-camera activity prompt
• Vision driver aihub — snapshot prompts through the hub, fallback to mock
• 13 providers — Gemini, GPT, Claude, Grok, DeepSeek, Groq, Ollama, OpenRouter, Azure, Together, Fireworks, Perplexity

---

🚀 Key Feature Highlights (Aether Cam Vision OS v2.6)

🎛️ Live Surveillance Matrix Wall

• Multi-Stream CCTV Grid: Toggle between 1x1 Focus View, 2x2 Wall, and 3x3 Security Matrix.
• Dynamic HUD Overlays: Real-time 30 FPS stream metrics, timestamp telemetry, and AI bounding box overlays.
• Vision Modes: Toggle Thermal Infrared Heat Filter and Green-Phosphor Night Vision directly on live canvas feeds.
• Simulated PTZ Control: Pan, tilt, and zoom controls directly integrated into camera cards.
• Instant AI Triage: 1-click anomaly evaluation directly from individual camera feeds.
• Dedicated full-page matrix available at /admin/live-camera-matrix.

🗺️ Zone Floor Plan & 2.5D Spatial Occupancy Map

• Pod-Level Seat Tracking: Real-time desk occupancy visualization for facility operations.
• Spatial Status Indicators: Distinct visual states for Verified Present (Green), Available Vacant (Gray), and Seating Anomaly / Mismatch (Red).
• Interactive Seat Inspector: Click any workstation to view assigned employee, department, presence status, and biometric confidence.
• Accessible on the landing page and via the Filament admin at /admin/floor-plan-explorer.

🧪 Interactive AI Vision Sandbox

• Multi-Model Triage Engine: Dispatches edge frames against custom vision prompts powered by imrandevbd/laravel-ai-hub.
• Pre-configured Edge Scenarios:
  • 🦺 PPE Safety Gear Check: Helmet & high-vis vest compliance auditing.
  • 🚪 Turnstile Tailgating Alert: Multi-person vestibule pass-through detection.
  • 🚨 After-Hours Perimeter Breach: Motion and unauthorized presence alerts with automatic lockdown escalation.
  • 🪑 Desk Occupancy Verification: Biometric seat match verification.
• Live Latency Telemetry: Sub-50ms inference simulation with structured JSON audit output.

🔑 1-Click Instant Role Switcher

• Fast demo access to Filament Command Center across 6 tailored permission tiers without entering passwords:
  • Super Admin (super@cms.test): Unrestricted master access across all sites, cameras, audit ledgers, and AI Hub Studio.
  • Facility Admin (admin@cms.test): Full management of buildings, zones, cameras, and employees.
  • Security Officer (security@cms.test): CCTV monitoring wall, tailgating flags, unauthorized perimeter alerts.
  • Operations Lead (ops@cms.test): Heatmaps, seat allocations, AI Copilot, and facility metrics.
  • HR Manager (hr@cms.test): People directory, biometric onboarding, daily executive briefings.
  • Executive Viewer (viewer@cms.test): Read-only executive dashboard and shift summaries.

📡 Real-Time APIs & Simulation Suite

• GET /api/system-telemetry: Live edge camera health, seat occupancy rate, open alerts count, and inference latency.
• POST /api/simulate-live-event: Dispatches realistic edge payloads to test asynchronous queues, anomaly detection, and movement logging.
• POST /api/ai-vision-sandbox: Real-time multi-model vision evaluation endpoint.

---

What you can do today

1. Site registry

Map a building the way it actually works: building → floor/zone → seat → camera.

Resource	What it stores
Buildings	Sites / offices
Zones	Floors, halls, rooms (floor + parent building)
Seats	Desk labels and optional current occupant
Cameras	RTSP URL, stream key, IP, vendor, online/offline, desk vs hallway, detection + face-match toggles, optional AI prompt
Employees	Code, department, expected seat, face photo, face-registered flag

Use cases:

• Desk cameras vs hallway / lobby cameras
• Assign an employee to an expected seat
• Turn detection or face matching off per camera
• Attach a custom AI prompt to a specific camera

2. Live operations dashboard (/admin)

Filament admin at http://localhost:8000/admin

• Camera health — online / offline / detection-enabled counts
• Live occupancy — seats with an occupant vs empty
• Attendance today — check-ins, check-outs, people still out
• Open anomalies — unresolved alert feed

Navigation groups: Registry, Monitoring, Attendance, AI, Reports.

3. Camera event pipeline

When a person (or motion) is detected:

1. Frigate posts to POST /webhooks/frigate (or you run the simulate command).
2. A raw camera_events row is stored (never process inline).
3. ProcessCameraEvent runs on the events queue.
4. Face matching returns employee_id + confidence (mock or CompreFace).
5. A business movement_logs row is written (enter / exit / present / idle / away).
6. If the camera has a seat, the occupant is updated.
7. Anomaly detection compares the camera zone/seat to the employee’s expected seat.
8. If the camera has an AI prompt, a vision job may run (rate-limited).

Frigate-shaped payloads (after.camera, after.label, after.snapshot.path) and a simple JSON body both work. The camera can be identified by id, name, or stream key.

4. Face recognition

• Upload a face photo on an employee → RegisterFaceEmbedding job → embedding row + face_registered = true
• On each event, FaceRecognitionService matches the snapshot
• Mock driver: uses employee_code in the webhook, or a filename like snapshots/E1001.jpg. unknown in the hint stays unmatched. FACE_RECOGNITION_FORCE_EMPLOYEE_ID can pin a match for demos.
• CompreFace driver: HTTP recognize/enroll client (set FACE_RECOGNITION_DRIVER=compreface)

Unmatched people on a camera still create a movement log and can raise an unauthorized-zone alert.

5. Anomaly detection and alerts

AnomalyDetectionService flags:

• Person not matched to any employee
• Matched employee in the wrong zone
• Matched employee on the wrong desk camera (seat mismatch)

Alert types already in the schema:

Type	Typical meaning
unauthorized_zone	Wrong place, or unmatched person
idle_too_long	Vision prompt returned idle / away
camera_offline	Health / ops (seeded as an example)
absent	Daily rollup / attendance gap

Severity: info, warning, critical. Alerts can be resolved in Filament (stores resolved_at + resolved_by).

6. Custom AI vision prompts

Configure prompts in Filament (name, text, model gpt-4o or gemini-1.5, applies to camera/zone, trigger on_event or scheduled).

• Only sampled / flagged events — never every frame
• Rate limit: one vision call per camera per AI_PROMPT_RATE_MINUTES (default 10)
• Mock driver returns working / idle / away from payload or filename hints
• OpenAI and Gemini driver shells are ready (AI_PROMPT_DRIVER=openai or gemini)
• Results stored in ai_prompt_results; idle/away also opens an alert

Example prompt: “Is this person actively working at their desk or away/idle?”

7. Attendance (independent of cameras)

Biometric / RFID / face terminals push or pull into attendance_records.

• POST /webhooks/attendance — check-in / check-out by employee code + device name
• Same-day punches upsert onto one record
• SyncAttendanceDevice hourly pull for devices without push
• Drivers: mock (demo punch), zkteco and hikvision stubs for vendor SDKs

Methods: fingerprint, RFID, face, PIN, webhook.

8. Daily reporting

Nightly AggregateDailyReport (01:15) writes employee_daily_summaries:

• Estimated time at seat
• Anomaly count
• Absence flag (no check-in that day)
• Raw movement counts (JSON)

The report table is what dashboards should read — not a live scan of millions of raw events.

9. Demo without hardware

Seeded demo (php artisan migrate --seed):

• HQ building, Operations + Lobby zones
• Seats A-12 / A-13 / A-14
• Cameras desk-a12, hall-3f, lobby-1
• Employees E1001–E1005 (some assigned seats, some face-registered)
• Sample movement, alerts, attendance, and a daily summary

php artisan cms:simulate-frigate-event desk-a12 --employee=E1001
php artisan cms:simulate-frigate-event lobby-1 --employee=E1002
php artisan cms:simulate-attendance E1001 --type=check_in

---

What this can grow into

The architecture is built for ~1,000 cameras as a control plane, not a video decoder.

Possibility	How
Real NVR	Point Frigate (or go2rtc + detector) webhooks at /webhooks/frigate. Split ~20–40 cameras per Frigate node.
Real face matching	Run CompreFace or a FastAPI + InsightFace service; set FACE_RECOGNITION_DRIVER=compreface. Laravel stays source of truth for employees.
Real vision QA	Flip AI_PROMPT_DRIVER to openai or gemini for “idle at desk?”, PPE, phone use, etc. Keep the rate limit.
Real attendance	ZKTeco / Hikvision push to /webhooks/attendance, or implement pull in the stub drivers.
Queue scale	Separate queues so a slow vision API cannot block face matches. Add Horizon on Linux.
Fast lookups	CameraLookupService caches camera → zone → seat in Redis.
Heavy history	Partition movement_logs by month, or move it to ClickHouse / Timescale.
Live occupancy wall	Laravel Reverb / Pusher (not required for this MVP; widgets poll).
Multi-site	Extra buildings/zones; same backend.
“Who isn’t working” reports	Read employee_daily_summaries, not raw events.
Offline camera alerts	Health probe / Frigate status → camera_offline alerts.
Scheduled prompts	ai_prompts.trigger = scheduled is modeled; wire a scheduler when needed.

Intentionally not in this app: RTSP decode, 24/7 frame analysis, Coral/GPU inference, fingerprint hardware, or storing raw video. Those belong at the edge.

---

Stack

Layer	Choice
App	Laravel 13, PHP 8.3+
Admin	Filament v4 (/admin)
Database	SQLite by default; MySQL 8 via Docker
Cache / queues	Database locally; Redis recommended in production
Queues	events, faces, vision, attendance, reports

Horizon is not installed on Windows (ext-pcntl). On Linux:

composer require laravel/horizon
php artisan horizon:install
php artisan horizon

---

Quick start

composer install
cp .env.example .env
php artisan key:generate
php artisan migrate --seed
php artisan storage:link
php artisan serve

Admin	http://localhost:8000/admin
Email	admin@cms.test
Password	password

With QUEUE_CONNECTION=sync (local .env), simulate commands finish immediately. For real queues:

php artisan queue:work --queue=events,faces,vision,attendance,reports

Optional MySQL + Redis

docker compose up -d

Then in .env:

DB_CONNECTION=mysql
DB_DATABASE=cms
DB_USERNAME=cms
DB_PASSWORD=cms
QUEUE_CONNECTION=redis
CACHE_STORE=redis

---

HTTP APIs

POST /webhooks/frigate

Header: X-Webhook-Secret when FRIGATE_WEBHOOK_SECRET is set.

Simple body:

{
  "camera": "desk-a12",
  "event_type": "person",
  "snapshot_path": "snapshots/E1001.jpg",
  "employee_code": "E1001",
  "detected_at": "2026-09-08T10:00:00+06:00"
}

event_type: person or motion. Also accepts Frigate’s after object.

POST /webhooks/attendance

{
  "device": "Lobby Reader",
  "employee_code": "E1001",
  "type": "check_in",
  "method": "fingerprint",
  "occurred_at": "2026-09-08T09:04:00+06:00"
}

type: check_in or check_out. method: fingerprint, rfid, face, pin, webhook.

---

Configuration

Variable	Purpose	Default
FACE_RECOGNITION_DRIVER	mock / compreface	mock
FACE_RECOGNITION_FORCE_EMPLOYEE_ID	Force a match (demo)	—
COMPREFACE_URL / COMPREFACE_API_KEY	CompreFace client	—
AI_PROMPT_DRIVER	mock / openai / gemini	mock
AI_PROMPT_RATE_MINUTES	Vision call cooldown per camera	10
OPENAI_API_KEY / GEMINI_API_KEY	Real vision APIs	—
ATTENDANCE_DRIVER	mock / zkteco / hikvision	mock
FRIGATE_WEBHOOK_SECRET	Frigate HMAC/shared secret	empty = open
ATTENDANCE_WEBHOOK_SECRET	Attendance webhook secret	empty = open
CMS_LOOKUP_CACHE_SECONDS	Camera lookup cache TTL	300

See config/cms.php.

---

Project map

app/
  Models/            Domain tables
  Services/          Face, AI prompt, anomaly, attendance, camera lookup
  Jobs/              ProcessCameraEvent, RunFaceMatch, EvaluateAiPrompt,
                     RegisterFaceEmbedding, SyncAttendanceDevice, AggregateDailyReport
  Http/Controllers/Webhooks/
  Filament/Resources + Widgets
  Console/Commands   cms:simulate-frigate-event, cms:simulate-attendance

Scheduled (when schedule:run / a cron is active):

• Hourly: pull attendance devices
• Daily 01:15: aggregate yesterday’s summaries

php artisan test

---

Suggested rollout

1. Registry CRUD + a handful of cameras (done)
2. Attendance webhooks (done; plug a real device)
3. One Frigate node → camera_events (done; point a real NVR at the webhook)
4. CompreFace / InsightFace → live movement_logs
5. Anomaly rules + alert routing (email/Slack can sit on alerts)
6. Paid vision prompts last (most expensive)
7. Dashboards on employee_daily_summaries only

Validate at 5 cameras before 500. The pipeline design holds at 1,000; schema and rate limits are what you want to get wrong early.

---

🤝 Let's Build Something Exceptional

I'm actively open to:
Remote Senior Full-Stack Roles · Freelance Contracts · Technical Partnerships · Long-Term Collaborations
in Laravel · WordPress · React/Next.js · AI-powered Platforms · Security Audits · SaaS Architecture

📍 Timezone: UTC+6 (Dhaka/Rangpur) — flexible overlap for US, EU & Asia
⚡ Available: Immediately · Production-first · Fast delivery · Transparent communication

Platform	Link
🌐 Portfolio	imrandev.bd
💼 LinkedIn	linkedin.com/in/imranbru99
🐙 GitHub	github.com/imranbru99
🐦 X / Twitter	@imrandev_bd
📺 YouTube	@ImranDevBD
📸 Instagram	@imranbru99
📘 Facebook	ExpertImranDev
🎵 TikTok	@imrandev_bd
🧵 Threads	@imranbru99
📌 Pinterest	@imrandev_bd
💬 WhatsApp	+880 1576-918420
📧 Email	me@imrandev.bd
🔗 All Links	linktr.ee/ExpertImranDev

"Security isn't an add-on — it's the foundation. Scale, speed, and trust drive every line of code I write."
— Imran Ahmed