FireEye Smart System

• The system supports fire detection through cameras.
• Alerts are sent via sound and messages to connected devices.

Dataset

DATASET REPO

$\color{red}{\textsf{Dataset Composition:}}$

a) Frames extracted and annotated from approximately1200 videos.

b) Images manually downloaded from the internet.

c)4312images sourced from various public datasets.

From a) and b) i collected 22970 images.

$\color{red}{\textsf{Repartition of scenes in dataset (22970 in total)}}$

The dataset is hosted on the Roboflow platform and consists of two projects. This dataset is public and can be downloaded and utilized for your own applications if desired. The links to the projects are: FireSmokeDataset_part1 and FireSmokeDataset_part2.

Dataset details

FireSmokeDataset_part1

FireSmokeDataset_part2

The distribution of the dataset for Training, Validation, and Testing tasks is illustrated in the figure below.

Model

• Training the YOLO model (v8s, v10m):
  • YOLOv8s prioritizes detection speed for camera devices using CPUs.
  • YOLOv10m prioritizes accuracy in detecting fire and smoke.
• The training process is monitored on W&B.
  • https://api.wandb.ai/links/giahoang481-h-c-vi-n-c-ng-ngh-b-u-ch-nh-vi-n-th-ng/jhq0xx2j

Demo Videos

https://github.com/HoangLayor/FireEye-Smart-System/blob/main/results/v2.0/video.gif

Reference:

• https://github.com/CostiCatargiu/NEWFireSmokeDataset_YoloModels