/FireDetectionCV

Computer Vision program for detecting fire in video and images.

Primary LanguageJupyter Notebook

FireDetectionCV (SST - Datathon)

Computer Vision program for detecting fire in video and images. The program was developed by the AIturbo team as solution of a case from Smart System Technologies in ML-fest hackathon.

About the project

  1. The fire detection training model was written in Python using Tensorflow & Keras;
  2. train_imgs dataset was used as data for training the model;
  3. PandNdivider.py script was used to divide images into 2 datasets: negative & positive by the labels in train_labels.csv file;
  4. Generated model name is fire_model.h5;
  5. fire_keras.py is the main program that processes all images in the directory test_imgs by our trained model and shows the results of testing;
  6. test_labels.csv contains results of the processing of all images of test_imgs dataset in which the class 1 means the presence of fire, and 0 its absence;

Testing & Results

Final results of testing our program in the test_labels.csv file!

Testing Results

Experiments

  1. We tried to use Haar cascades for fire detection, and trained the fire_cascade.xml, but unfortunately the accuracy was very low.
  2. Finally we decided to use Tensorflow & Keras technologies for training our model, using Tensorflow.js library.

Acknowledgments