This study presents a comparative analysis of YOLOv3 and SSD architectures for real-time object
detection in surveillance systems, focusing on key performance metrics such as mean Average Precision
(mAP), Intersection over Union (IoU), and Frames Per Second (FPS). Utilizing datasets such as COCO,
AI City, and PETS, the results reveal that YOLOv3 outperforms SSD in terms of speed, achieving nearly
double the FPS, making it more suitable for real-time applications where low latency is critical. While
both models demonstrate strong performance in object detection and localization, YOLOv3 consistently
shows higher mAP and IoU values, indicating superior accuracy and precision in diverse surveillance
scenarios. These findings underscore the effectiveness of YOLOv3 in real-time surveillance, while SSD
remains a competitive option when slightly higher accuracy is required despite a trade-off in processing
speed.
