A Comparative Study of Deep Learning Architectures for Real-Time Object Detection in Smart Surveillance Systems

Smart surveillance systems have increasingly leveraged deep learning for real-time object detection, enhancing public safety, traffic monitoring, and anomaly detection. This study compares leading deep learning architectures—YOLOv5, SSD, and Faster R-CNN—for their performance in smart surveillance contexts. The models are evaluated on standard metrics such as accuracy (mAP), speed (FPS), and computational efficiency using the COCO and custom surveillance datasets. Our findings indicate that while YOLOv5 excels in real-time performance, Faster R-CNN offers higher accuracy but lags in speed. SSD provides a balance between the two. This comparative analysis aims to guide the selection of appropriate architectures based on specific surveillance system requirements.