A Novel Approach for Surveillance Compression using Neural Network Technique

Abstract

The integration of closed-circuit television (CCTV) monitoring is crucial in the field of video processing, which provides an efficient method for comprehensive surveillance. However, a key challenge associated with this practice is its substantial demand for storage space. Typically, surveillance footage is stored in hard disk drives, and due to limited storage spaces, it is deleted after some time. To address this issue, an innovative method for compressing CCTV video, named object detection-based surveillance compression (ODSC), is introduced. Our ODSC model is divided into two steps: -i) depending upon the objects in the video, determine the significant and non-significant frames of surveillance video using the neural network approach YOLOv5s & YOLOv7-tiny and Yolov8s ii) construct the video of significant frames. Following a comprehensive analysis of the experimental outcomes, it is noted that YOLOv8s stands out with a remarkable detection accuracy of 99.7% on the COCO dataset. Our ODSC approach is reducing the storage space greatly and achieving an average compression ratio of up to 96.31% using YOLOv8s, which surpasses the existing state-of-the-art methods.