Computer vision2019

Fire & Smoke Vision Prototype

Computer-vision experiments and a monitoring interface for detecting visual signs of fire and smoke.

My contribution

Software development, vision experiments and embedded integration

Watch the project Discuss related work

Project archive

Annotated Windows monitoring interface: live stream, device state and event history.

ENGINEERING HIGHLIGHTS

What stands out.

Vision & neural-network prototype
Hardware I/O, GSM & operator alerts
Embedded device, web & desktop interfaces

Project overview

From the problem to the implementation.

The problem

Camera-based monitoring needs to connect visual detection experiments with a clear interface for status, events and alarms.

The approach

Explored OpenCV image-processing methods and TensorFlow/Keras models.
Worked with image annotations and dataset-preparation tooling.
Developed monitoring components for camera status, event handling and alarms.

What came together

A prototype workspace connecting vision experiments, embedded-camera integration and an operator-facing monitoring application.

A closer look

Inside the project

37 images

Enclosed prototype internal assembly — v1-hardware-inside.

Original device schematic illustrating the embedded components.

Front view of the enclosed prototype; no certification is implied.

Enclosed prototype internal assembly — Ver1-Inside-1.

Enclosed prototype internal assembly — Ver1-Inside-2.

Enclosed prototype internal assembly — Ver1-Inside-3.

Enclosed prototype internal assembly — Ver1-Inside-4.

Enclosed prototype internal assembly — Ver1-Inside-5.

Enclosed prototype internal assembly — Ver1-Inside-6.

Enclosed prototype internal assembly — Ver1-Inside-7.

Enclosed prototype internal assembly — Ver1-Inside.

CCTV-style prototype internals and custom electronics.

Annotated Windows monitoring interface: live stream, device state and event history.

Fire-detection device web control panel — screen 1.

Archived web configuration view showing emergency-output and SMS settings.

Fire-detection device web control panel — screen 9.

Fire-detection device web control panel — screen 10.

Fire-detection device web control panel — screen 11.

Fire-detection device web control panel — screen 12.

Fire-detection device web control panel — screen 13.

Original Fire & Smoke Vision Prototype project banner.

Project footage

See it in action

OpenCV fire test · 02

Second OpenCV fire-test recording. Silent demonstration.

Open video

OpenCV fire test · 01

Physical fire-test recording — OpenCV stage. Silent demonstration.

Open video

Simulated detection workflow

Simulated demonstration — fire-detection workflow. Silent demonstration.

Open video

Indoor detection test

Indoor fire-detection test recording. Silent demonstration.

Open video

System view

How the system works

Camera input

Pi Camera supplies imagery close to the embedded processing unit.

Candidate regions

C++/OpenCV methods identify image regions that may contain fire or smoke.

Second-stage analysis

A neural-network stage evaluates candidate imagery.

Device event logic

Python coordinates status and alarm-related events.

Alert channels

Hardware I/O, SMS/GSM and local-network interfaces expose events.

Operator tools

Windows/PyQt monitoring and the HTTP panel provide live imagery, status and settings.

Technical notes

The detail behind the build

A two-stage vision pipeline

C++ and OpenCV identify image regions that may contain fire or smoke. A neural-network stage then examines candidate imagery. The prototype brings these detection stages together with an embedded device and an operator interface.

Processing close to the camera

A Raspberry Pi provides the processing unit near the camera. The hardware design brings camera input, digital I/O and network communication into a compact assembly intended for installation alongside the monitored scene.

From detection to operator awareness

Detection events feed several output paths: hardware I/O, SMS/GSM, a local browser view and a Windows application. The desktop interface combines live imagery, network/device status and event records.

Live camera view

Pre-alarm and alarm events

Device status and event history

Database connection settings

A browser interface for device settings

A local HTTP interface exposes configuration and monitoring without a dedicated desktop installation. Its settings cover device status, emergency outputs and SMS-related functions.

Hardware development and testing

The project includes an enclosed prototype and a CCTV-style assembly with GSM connectivity. Internal hardware photographs, a schematic, interface screenshots and demonstration recordings show how the vision software and physical device were brought together.

Implementation

Software, hardware & tools.

Software

C++ vision processing & neural-network experiments
Python desktop & web monitoring

Hardware

Raspberry Pi
Pi Camera
Custom electronics
SIM800 GSM module
Enclosure variants
Hardware alarm outputs

Technology stack

Computer vision
Embedded integration
Alert workflows

Continue the conversation

Working on a related challenge?

Get in touch about Python, embedded systems or a research opportunity.

Get in touch View résumé

Next project

Embedded QR Scanning Appliance

Embedded systems