Design and Implementation of Smart Surveillance System Using Deep Learning Method

Abstract

This works explored deep learning techniques to create a Smart Surveillance System, also known as Smart CCTV, for vehicle categorization. Unlike traditional Smart CCTV systems, which rely on centralized video processing and analysis, this system is self-contained and capable of producing textual data on its own. To develop the model, the YOLOv7 algorithm method is employed, and the process of creating datasets, labeling, training, verifying, and testing the deep learning model is detailed. To ensure platform compatibility, the trained models are executed on the Jetson Nano device, and the whole procedure and output data processing are performed in Python. We also performed durability testing on various weather condition. The proposed system is intended to provide a more efficient and dependable surveillance system for applications such as traffic monitoring and security. Where can devices that are independent in terms of process and data processing be obtained, so that output is obtained in the form of data on the number and classification, as well as the status and performance of the devices being built. The developed series of devices perform well in terms of status and performance. CPU utilization is within acceptable limits, and physical RAM consumption is consistent. Despite an increase in SWAP RAM consumption, the stable fan speed indicated that there was no significant overheating.