Video Compression By Eliminating Irrelevant Frames

Abstract

Regarding the security of the ATM systems situated at various locations throughout the world, constant monitoring via surveillance at the ATM machine is required. Even though ATMs are equipped with manual protection, several theft events have made headlines over the past few decades. Since continuous surveillance provides an alternative to manual security, it is not feasible for the banks to do so at every one of these ATMs. With so much data generated by the monitoring system inside and outside the ATM, it becomes expensive for the banks to manage in terms of memory. As a result, the goal of this job is to reduce the number of frames in the video that was collected by removing all of the pointless frames. The technique employs transfer learning on a Mask RCNN Network to identify frames containing people before removing any such frames. Mask RCNN network, which is typically trained on the COCO dataset, will be used to detect humans in the so-called relevant frames and subsequently to filter out the frames with no humans. The pertinent frames are after that turned into video and kept in the backup. The plan decreases the amount of storage needed for continuous monitoring. Keywords- Region-Based Convolutional Neural Network (RCNN), Convolutional Neural Networks (CNN), Closed Circuit Television(CCTV), Discrete cosine transform (DCT), GPU (Graphics Processing Unit), Deep Learning techniques, area classification,