View Proposal

---

Proposer

Medhdhar Salem Ali Muthanna Al- Gaashani

Title

Fire and Smoke Detection from Images Using Deep Learning

Goal

Develop a deep-learning object detector that marks fire and smoke regions in images or short video frames, supporting early safety monitoring in buildings, laboratories and public areas.

Description

Early detection of fire and smoke in camera footage can prevent damage and save lives. This project builds an object-detection prototype that draws bounding boxes around fire and smoke regions. Students use a public bounding-box dataset, convert labels to YOLO format if needed, and train a lightweight detector such as YOLOv8n on two classes: fire and smoke. The detector is evaluated using precision, recall, and mean average precision (mAP). A Streamlit or Flask demo processes an uploaded image or video frame and displays the detected boxes. False alarms (sunlight, orange objects, fog) are analyzed as failure cases.

Resources

D-Fire dataset (~21k YOLO-labeled fire/smoke images): https://github.com/gaiasd/DFireDataset — Example YOLOv8 + D-Fire + Streamlit implementation: https://github.com/AlimTleuliyev/wildfire-detection

Background

Url

Difficulty Level

Moderate

Ethical Approval

None

Number Of Students

1

Supervisor

Medhdhar Salem Ali Muthanna Al- Gaashani

Keywords

deep learning, ml, ai

Degrees

Bachelor of Science in Computational Sciences and Software Engineering