Detecting Criminal Activities of Surveillance Videos using Deep Learning

Videos generated by surveillance cameras inside the ATM were very long. In case, any robbery had taken place inside the ATM; it became time consuming to watch the entire long video. Hence, there was a need to process these surveillance videos by extracting the priority frames from it in which suspicious activities like robbery, murder, kidnap, etc. had taken place. The objective of this paper was to propose algorithm that would generate a detect the suspicious frames from that long surveillance video for the authorities which would consists of priority information. In this paper a novel approach dealing with Convolutional Neural Networks using Deep Learning was used to sample the priority information from the surveillance videos. The priority information was the suspicious activities like robbery, murder, etc. which take place inside the ATM. The results of the CNN model effectively were able to extract suspicious activity frames from a long video and thus extract suspicious frames and create a video from it.