FRVC: frame relevance based video compression for surveillance videos using deep learning methods

The prevalence of closed-circuit television cameras (CCTV) has witnessed a rapid escalation over recent decades. CCTV serves the purposes of safety, security, and monitoring across various sectors, that contribute to smart infrastructure and sustainable urban development. Though the requirement for CCTV increases, it faces major challenges in surveillance video storage, particularly in high-resolution systems. This results in substantial storage demands, leading to economic and environmental concerns regarding resource utilization and management, aligning with the goals of resource efficiency and sustainable consumption. To address these issues, we propose a frame relevance-based video compression (FRVC) algorithm comprising three phases: (i) dataset preparation, (ii) relevance frame classification, and (iii) video compression. This approach ensures the resilient and efficient management of video data, which enhances its usefulness in urban monitoring. The FRVC framework was tested using a customized surveillance video dataset under three scenarios. In the relevance frame classification module, Mask region-based convolutional neural network (Mask R-CNN) and You Only Look Once version 9 (YOLOv9) object detection approaches are used to detect relevant frames of surveillance video where YOLOv9 surpasses Mask R-CNN in terms of evaluation metrics (accuracy, precision and F1-score) for 80–20 dataset ratio. The proposed FRVC achieves a compression rate up to 96.3%, 99.9% and 63.3% for three different scenarios respectively. Compressed videos maintain the same resolution and frame rate as compared to the original video. This innovative framework supports sustainable technology adoption in the development of surveillance systems, contributing to long-term data storage solutions and cost-effective urban monitoring.