logo Smart Monitoring System

A smart home monitoring system combining computer vision, voice AI, Bluetooth radar and Gemini visual recognition. This manual walks you through the five modes and every feature, step by step.

Live site: smart-monitoring-system.zeabur.app Recommended browser: Chrome / Edge Runs fully in the browser — no install
Main system QR
Smart Monitoring System
Mobile version QR
Mobile

Highlights

What sets this system apart from ordinary surveillance.

Real-time in the browser
AI recognition runs right in the browser — just open the page. No install, no GPU server.
Video stays private
Camera frames are processed on your device and never uploaded to the cloud — smart yet private.
Seven languages, instant switch
Traditional & Simplified Chinese / English / Korean / Spanish / French / Hindi — interface, alerts and voice all switch together.
Five modes in one place
Find Items, Leaving-Home Reminder, Cabinet Alert, Security Monitor, Smart Assistant.
Two devices working together
The Bluetooth scanner acts as a “finding radar”; the voice-radar hub handles “listening and speaking”.
Phone remote sync
Scan the QR to view the feed and change modes from your phone — two-way sync with the console.

Hardware

The system is made of two physical devices — a Bluetooth scanner for “finding things”, and a smart voice-radar hub for “listening and speaking”.

Smart Voice-Radar Hub
Smart Voice-Radar HubSmart Voice-Radar Hub
Bluetooth Scanner
Bluetooth ScannerBluetooth Scanner
Smart Voice-Radar Hub
The main unit. With built-in speaker and microphone, it handles the voice assistant’s “listening” and “speaking”, and announces radar search results and alerts.
Bluetooth Scanner
A directional antenna scans the signal strength of nearby Bluetooth tags to estimate an item’s direction and distance — the radar behind “Find Items”.

How the two devices reach the system

Both devices talk to the system over your home Wi-Fi. They are optional — every mode works from the browser alone; the hardware adds radar search and spoken replies.

  1. Power up both devices and put them on the same Wi-Fi network as the computer running the console.
  2. The scanner reports Bluetooth signal strength to the console, which turns it into the direction and distance shown in “Find Items”.
  3. The hub carries the voice assistant’s microphone and speaker, so radar results and alerts are read out loud in the room instead of only on screen.
TipThe system itself runs in the cloud while the devices sit on a local network, so a computer on that same network relays between them. If the hardware goes quiet, first check that this computer is awake and on the right network.

Architecture

How data flows: sensor input → real-time AI in the browser → multiple outputs.

INPUT
Camera feedLive video
Bluetooth scannerSignal strength → bearing
Voice-radar hubMicrophone input
CORE
Real-time AI inference in the browser
YOLO object detection (11 specialized models) Face + age recognition Pose + gesture analysis Gemini Vision API
YOLO objects · faces · pose · Gemini | video never leaves the device
OUTPUT
Live alerts
Video records
Phone / Telegram notify
Voice announce
PrivacyAll image recognition happens in your browser (on-device); frames are never uploaded to a cloud server.

Quick Start

Up and running in three minutes: open the site, allow the camera, pick a mode, press Start.

overview
Fig. 0: Home overview — mode bar on top, settings on the left, live monitoring in the center, alerts / clips / analytics on the right.

Signing in

The system requires an account. A sign-in screen appears the first time you open it — you can register yourself, no application needed.

  1. No account yet: tap “Register” under the sign-in box, choose a username and password, then press “Register & Log In” — you go straight into the system.
  2. Already have an account: enter your username and password and press “Log In”.
  3. After signing in it asks where to save event clips. Pick a folder on this computer, or press “Skip for now” and set it later under Settings → System → Storage.
TipSign in on the phone with the same account — that is what keeps modes and alerts in sync with the console. Faces, items, alert history and clips are kept separately per account.

Getting started

  1. Open the system URL in a browser on your computer or phone and sign in as described above.
  2. Wait for the recognition models to load (bottom-right). The first time downloads the models; afterwards they’re cached.
  3. When the browser asks, click “Allow” to enable the camera.
  4. Pick the mode you want from the mode bar at the top.
  5. After setting the left-column options, press Start to begin monitoring.

Three main areas

Left · Mode settings
Options specific to the current mode — which items to track, which to protect, the cabinet area, and so on.
Center · Live monitoring
The live camera feed with detection boxes, face boxes and a status bar overlaid. Dual cameras supported.
Right · Alerts / analytics
Live alert list, disappearance-clip playback, and the item social-network graph.
TipOn first login an interactive Guide opens automatically, spotlighting the key controls of the current mode. Reopen it anytime with the “Guide” button up top — its steps follow whichever mode you’re in.
TipThe whole system supports seven languages (Traditional & Simplified Chinese / English / Korean / Spanish / French / Hindi), switchable under “Settings → System → Language”. Interface text, alerts and voice all change with it.

2 Leaving-Home Reminder

When a registered face is recognized, it reminds the person “what they forgot to take”, and can check the weather to decide on an umbrella.

leaving home
Fig. 2: Leaving-Home Reminder — must-take list, city weather and face registration.

Steps

  1. First register family faces under “ Face Registration” on the right (see section 9).
  2. Tick the must-take items for going out (keys, wallet, phone, umbrella…).
  3. Select your city to get weather information.
  4. Press Start. The reminder fires only when a registered face leaves the frame while a must-take item is still there — it won’t nag while you’re home.
TipWhen something’s missing it warns “ So-and-so forgot to take ○○”.
TelegramWhen this mode detects the registered face and flags missing items, it automatically sends a Telegram notification to the user.
telegram leaving home
Fig. 2-1: Leaving Home Telegram alert — Camera 1 carries the message, Camera 2 follows with the other angle.

3 Cabinet Alert Mode

Monitors cabinet doors and judges violations by “age” or “face” rules (e.g. children may not open the medicine cabinet).

cabinet alert
Fig. 3: Cabinet Alert mode — box the cabinet area on the image and set opening rules.

Steps

  1. Click “ Cabinet Alert” at the top.
  2. Click “Capture current frame” to grab a still from the camera.
  3. Click two opposite corners on the image to box the cabinet area to monitor.
  4. Set the opening rule: an allowed age range, or specific faces.
  5. Press Start. If the door is opened against the rule, it raises a red alert at once and auto-captures the moment.
CautionYou must box the cabinet area first; only then does the system start detecting door state, avoiding false alarms.
TelegramWhen an unauthorized cabinet opening is detected, the system automatically sends a Telegram notification to the user.
telegram cabinet alert
Fig. 3-1: Cabinet Alert Telegram message — when a minor opens the cabinet, both camera views are sent to your phone.

4 Security Monitor Mode

Protect chosen items against grabbing, detect falls, and frame a child/pet danger zone that alerts when a person or pet enters and stays.

security monitor
Fig. 4: Security Monitor mode — tick protected items and features (grab / fall / danger-zone detection).

Steps

  1. Click “ Security Monitor” at the top.
  2. Tick the items to protect.
  3. Tick the features to enable: grab detection, fall detection, danger-zone alert.
  4. (Optional) Set identity rules: an alert fires only when someone outside the allowed ages / faces makes a grab gesture.
  5. (Optional) After ticking “Danger Zone Alert”, press “Danger Zone”, capture the frame and click two corners to mark the area; limit targets to people / pets or “young children (age 1–15)”, and set the dwell time.
  6. Press Start. An alert fires when a protected item is grabbed, a fall is detected, or a target enters and stays in the danger zone.
TelegramWhen a protected item is taken, a fall is detected, or someone enters the danger zone, the system automatically sends a Telegram notification.
telegram security
Fig. 4-1: Security Monitoring Telegram message — the moment a protected item is grabbed, plus who did it.

5 Smart Assistant Mode

Analyzes your carrying habits and offers a social-network graph, loss-risk prediction, carry recommendations and item-location lookup.

assistant
Fig. 5: Smart Assistant mode — item social network, loss risk, carry recommendations, location lookup.
Item social network
Shown as item tags plus pair cards: the most-carried items and the pairs most often carried together (each with a strength bar and count) — clear at a glance, no tangled graph.
Loss-risk prediction
Gemini reasons about which items are easy to forget or misplace.
Carry recommendations
Recommends items “you might also want to take” for the situation.
Item-location lookup
Look up where a given item appears most often.
At a glance
Summary cards showing the highest-risk item and the latest recommendation the moment you enter the mode.
social network
Fig. 5-1: Item social network — item tags for the most-carried items (with counts) on top, and cards for the pairs most often carried together (with strength bars and counts) below.

6 Live Monitoring

The center camera feed with all recognition results overlaid.

live monitoring
Fig. 6: Live monitoring — detection boxes show item name and confidence; moving items show speed.
  • Detection boxes: only your selected items are boxed, labeled with name and confidence.
  • Motion trail: each item’s path is drawn on screen and stays visible, so you can trace where it went.
  • Dual cameras: if a second camera is connected, both show side by side.
  • Status bar: shows the current state (monitoring / alert).

7 Alerts

Lists the alerts the system raises in real time, color-coded by level.

InfoGeneral notices, e.g. “all items present”, “cabinet opened”.
WarningWorth attention, e.g. “item disappeared”, “a hand is near a protected item”.
DangerNeeds immediate action, e.g. “cabinet violation”, “fall detected”, “danger-zone intrusion”.

Alert text follows your chosen language; the list can be cleared at any time.

Alert Log

The alerts area keeps a log of recent alerts, one row each as “time + colored dot + message” — as easy to read as a phone's notification list, so you can review what happened without opening anything. The log is stored in this device's browser.

History Center

Tap “🕑 History” at the bottom-right to open a full alert timeline: each alert carries a snapshot thumbnail, and you can filter by mode (Leaving-Home / Cabinet Alert / Security Monitoring), search by keyword, or export CSV in one tap. History is kept per account on the local device.

8 Settings — Registration

Register faces and items for identity checks used by Leaving-Home, Cabinet and Security modes.

registration
Fig. 8: Settings → Registration tab — registered people, face registration, item registration.

Face Registration

  1. Upload a photo, or take one live from the main camera.
  2. Enter the name.
  3. Click “Register Face”.

Item Registration

  1. Upload one or more photos of the item.
  2. Choose the matching item category.
  3. Click “Register Item” (it links to the name entered in face registration).

Home map

Settings → Home map lets you sketch your floor plan and pin each item where it belongs (keys by the entrance, pillbox on the dining table). One set of pins feeds two features.

  • Item map: open it from the Item Search toolbar to see where each item was last seen, and which camera is nearby.
  • Misplaced alert: warns you when an item sits somewhere it does not belong. It judges by these pins, so items you never pinned never trigger it.

9 Settings — System

Adjust detection source, audio, time zone, language, appearance, cameras, account and video save location.

system settings
Fig. 9: Settings → System tab — category nav on the left, matching settings on the right.
Detection sourceToggle WebGPU-accelerated recognition (faster when the device supports it) and fine-tune per-item detection confidence.
Audio outputChoose which speaker the voice announcements use.
Time zoneSet the time zone for clips and records (default Taiwan UTC+8).
Language繁體中文 / 简体中文 / English / 한국어 / Español / Français / हिन्दी — switches the whole interface instantly.
AppearanceSwitch between dark and light themes, applied instantly.
CamerasChoose the device source for Camera 1 / Camera 2.
AccountShows the signed-in account and role, and lets you change the password.
StorageChoose the local folder for event clips (videos stay on this device only, never uploaded).
AboutShows the current system version.

10 Phone View

Scan the QR code with your phone to see the same live feed as the console.

phone qr
Fig. 10: Phone view — scan this QR code with your phone to open the mobile interface.
  1. Click the “ Phone Viewer” button at the top to open the QR window.
  2. Scan the QR code with your phone camera; on a computer you can also click the QR code to open the viewer site directly.
  3. Log in with the same account as the console, and your phone shows the live feed; modes and settings sync both ways.

Phone UI — Main screen

Once connected, the phone shows a compact monitoring UI: alert status with clear/pause buttons on top, the live camera feed in the middle (switch Camera 1/2, take a snapshot), and System Control, “5-seconds-before” clips and the alert log below.

phone main
Fig. 10-1: Phone main screen — alert status, live camera, system control, clips and alert log.

Phone UI — System Control (Adjust)

Tap Adjust under “System Control” to switch the operating mode and tick tracked items right from your phone; press “Apply” and it syncs back to the main system at once — no need to walk back to the computer.

phone control
Fig. 10-2: Phone system control — switch mode, pick tracked items; two-way sync after Apply.

Phone UI — Settings

Tap the in the top-right to open settings, where you can switch between dark/light themes, see the signed-in account and log out.

phone settings
Fig. 10-3: Phone settings — theme switch and account.
TipThe phone UI and the console are two-way synced: change the mode or tracked items on the phone and the main system follows; the main system’s alerts are also pushed to the phone in real time.

11 Voice Assistant

Wake it by voice and give commands to operate the system hands-free.

Tap “Smart Assistant” up toplistening starts automatically
Say "Smart Assistant"Hear "I'm listening"
Give your commandSpeech recognition
System replies by voiceand executes the action

Steps

  1. Press “Smart Assistant” at the top to enter the mode; the status shows “Listening”. Voice capture defaults to the voice-radar hub’s mic — switch to this computer’s microphone under “Settings → Audio”.
  2. Say the wake word “Smart Assistant” and hear “I’m listening”.
  3. Then give a command, e.g. “Find my keys”, “Switch to Security Monitor”, “Check the weather”, “Turn off the alert”.

Example commands

"Find my keys"Switches to Find Items mode, tracks keys in real time and announces their location.
"Switch to Security Monitor"Switches the operating mode; say any of the five mode names.
"Check the weather"Reports the current city's rain probability and whether to bring an umbrella.
"Turn off the alert"Clears all alerts and silences the alert sound.
"What should I take today"Gives carry recommendations based on your habits and reads them aloud.
TipSpeech recognition and read-out follow the interface language: choose English and it listens and speaks in English; same for Korean / Spanish.

12 Troubleshooting

The problems that actually come up, and how to clear them. Most are browser permissions or cache — not a broken system.

I pressed Start but the view is still blackThe browser asks for camera permission first — choose Allow. If no prompt appeared, click the padlock at the left of the address bar, set Camera to Allow, and reload the page.
The camera won’t open, or says it’s in useOne camera cannot be used by two programs at once. Close other video apps — and any other tab running this system — then press Start again.
Video works but no detection boxes at allGo to Settings → System → Detection and untick “GPU acceleration (WebGPU)”, then Stop and Start again. On some Chrome builds WebGPU miscomputes silently, and every box disappears at once.
No Telegram notifications arriveYou have to link it first: get the link under Settings → System → Notifications, open it on your phone and send /start in Telegram — it is done when the subscriber count reads 1. Note that Item Search never sends notifications; that is deliberate.
The voice assistant doesn’t wake upCheck that microphone permission is allowed. The wake word is “Smart Assistant”; wait for it to answer “I’m listening” before giving the command. In a noisy room, speak closer to the mic.
The phone shows no live viewThe phone must be signed in with the same account as the console, and the console must be running (the status bar reads “System Monitoring”). The console pushes the video — stop it and the phone goes blank.
After switching language some text stays, or the layout breaksThe browser cached an old file. Do one hard reload (Ctrl+Shift+R on Windows, Cmd+Shift+R on Mac). Once is enough.
I can’t find the clip that was just recordedEvent clips are saved only to the local folder you chose at sign-in — they are never uploaded. Check the path under Settings → System → Storage; if you pressed “Skip for now”, no folder has been set yet.

13 Privacy & your data

A surveillance system only earns trust if it is clear about where the data goes. Here is exactly what stays on your machine and what does not.

Camera framesRecognition runs inside your browser, on your own device. Frames are never uploaded. You can see where the computation happens under Settings → System → Detection.
Event clipsWritten only to the local folder you chose at sign-in, and never sent to the cloud. Delete the folder and the clips are gone.
Alert log and historyKept in the browser on the device you are using, separately per account. Clearing your browser data clears them.
Account and passwordPasswords are stored as a salted PBKDF2-SHA256 hash — the plain password is never kept, so nobody can read it back out of the system.
Registered faces and itemsThese are kept on the server, so the same account recognises the same people from any device. Only your own account can read them, and deleting an entry in Settings → Registration removes it.
Telegram notificationsSent only to the Telegram account you linked yourself. Nothing is broadcast, and you can unlink at any time under Settings → System → Notifications.
TipIn short: the video never leaves your device, and the only things that reach the server are what the system needs to recognise you again — your account and the faces and items you deliberately registered.