A Comprehensive Review on Real-Time Vehicle and Pedestrian Detection Using YOLO

Real-time vehicle and pedestrian detection has been considered one of the fundamental tasks of building an autonomous driving and intelligent transportation system. The current study is a comparative investigation into the performances of some of the state-of-the-art models from YOLO (You Only Look Once), which spans YOLOv7 to YOLOv11. YOLO is a real-time object detection algorithm that detects objects in an image using a single neural network evaluation and predicts the bounding boxes and class probabilities. These models were checked using the COCO dataset on real-time object detection, in an architecture incorporating anchor-free detection, further pushed forward by Vision Transformers, enabling good accuracy, efficiency, and improved inference speed. Our results show consistent improvement over the revisions, where the best performance of YOLOv11 yields 88% MAP (Mean Average Precision) and 45 FPS (Frames Per Second), outperforming previous versions of YOLO and state-of-the-art works in the literature. Comparisons with other existing methods reveal that improvements are indeed effective, especially with Dynamic Inference Scaling to ensure adaptability across a wide range of hardware platforms. These findings position YOLOv11 as a robust and scalable solution for real-time vehicle and pedestrian detection, paving the way for further advancements in safety-critical applications.