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.
- The fire detection training model was written in Python using Tensorflow & Keras;
train_imgs
dataset was used as data for training the model;PandNdivider.py
script was used to divide images into 2 datasets: negative & positive by the labels intrain_labels.csv
file;- Generated model name is
fire_model.h5
; fire_keras.py
is the main program that processes all images in the directorytest_imgs
by our trained model and shows the results of testing;test_labels.csv
contains results of the processing of all images oftest_imgs
dataset in which the class1
means the presence of fire, and0
its absence;
Final results of testing our program in the test_labels.csv
file!
- We tried to use Haar cascades for fire detection, and trained the
fire_cascade.xml
, but unfortunately the accuracy was very low. - Finally we decided to use Tensorflow & Keras technologies for training our model, using Tensorflow.js library.