Machine learning based approach for multimedia surveillance during fire emergencies

Video based surveillance of manmade disasters such as fire has become very hot topic in research and it is playing an important role in the development of smart environment. The disasters like fire cause many economic and social damages. We can prevent these damages by early detection of the fire. The current advancement in embedded processing have permitted the detection of fire using vision-based i.e. Convolutional Neural Networks (CNNs) for the surveillance. Therefore, we proposed a method using machine learning techniques for Multimedia Surveillance during fire emergencies. Our proposed model has two main deep neural networks models. Firstly, we used a hybrid model made of Adaboost and many Mulit-layer perceptron (MLP) neural networks. The purpose of hybrid Adaboost-MLP model is to predict fire efficiently. This model used different sensors data like smoke, heat, and gas for training. After predicting the fire, we proposed a CNN model to detect the fire immediately. These results show that our trained model has near 91% fire detection accuracy. We can the false positive results are quite low. These results can be improved more by further training.