Comparative Study of Object Detection Algorithms: YOLO, SSD, and Faster R-CNN

Abstract

Object detection has become a cornerstone task in computer vision, enabling applications in
surveillance, autonomous vehicles, and augmented reality. This paper compares three state-of-
the-art object detection algorithms—YOLOv3, SSD (Single Shot MultiBox Detector), and
Faster R-CNN—based on accuracy, speed, and resource utilization. We evaluate each model
on PASCAL VOC and MS COCO datasets, measuring metrics such as mean Average Precision
(mAP), inference latency, and frames per second (FPS). YOLOv3 achieves the fastest
performance (up to 45 FPS on a single GPU) with slightly lower accuracy compared to Faster
R-CNN, which delivers the highest mAP (79.2 on VOC) but at the cost of slower processing
(7 FPS). SSD offers a middle ground with competitive accuracy and moderate speed. We
analyze model performance across object scales, occlusion scenarios, and lighting variations.
YOLO performs best on large, centered objects, while Faster R-CNN handles small or partially
obscured objects more reliably. Resource usage tests on embedded devices suggest YOLO and
SSD are better suited for real-time edge applications. The findings inform practitioners
choosing detection algorithms for specific deployment constraints, balancing between speed
and precision. Our work provides a practical benchmark and guidance for deploying object
detection in real-world systems.