Publication

Abstract : In places where more number of people are present, it is difficult for people to escape that place during emergency situations. So, there arises a need to detect the severity of smoke as early as possible as smoke develops rapidly, and the officials must be alerted on the situation immediately. So, in this paper, we present a method that would make use of the camera feeds and take the images frame by frame to detect the presence and progress of smoke. We extract some features of smoke from the images. The features obtained are then fed into both of these models (LSTM, GRU). Both LSTM and GRU are mainly use the knowledge of previously obtained output over a period of time for prediction rather than just using only the output of the most recent layer like other RNN models. We use these two models as the data that is input to the model is a time series data. We discuss the advantages and compare the accuracy of the results obtained by using both LSTM and GRU methods. This system can be developed and used in smart spaces that would help for quicker response from officials. Since we use cameras that are already installed in a building, this eliminates the need to use sensors.