Items in eSource are protected by copyright. Previously published items are made available in accordance with the copyright policy of the publisher/copyright holder.
Setting up the surveillance camera system has become inevitable in all the places throughout the world. Huge number of human resources has been deployed for monitoring the surveillance system in which the unusual happens very rarely. Also, It requires manual operation to review the footage for any abnormal incidence. Taking this as a use case, this thesis work designated to provide the initial solution for the same with the machine learning techniques to avoid involving human resource for monitoring any abnormal activities which got identified in the broadcast video from the surveillance system.
To identify the abnormal activities, it requires to relate the video frames in the past and the current to predict for the unprecedented activities in the video frame. Considering the nature of the required machine learning model, Long Short-Term Memory (LSTM) has been chosen which is of a type recurrent neural networks (RNN). The basic feature of the chosen algorithm is to persist data for the future use and make any decisions by considering the information stored in the memory. Handling the temporal data is one of the main reasons behind choosing the LSTM algorithm. This paper aims at developing the solution, only to identify the explosions in the video and later shall be enhanced.
