ROBUST REAL-TIME 2D OBJECT DETECTION USING YOLOV5: ARCHITECTURE, TRAINING OPTIMIZATION, AND COMPARATIVE EVALUATION

Abstract

Real-time object detection is an essential task in computer vision applications such as autonomous driving, video surveillance, and robotics. This paper presents an in-depth study and implementation of YOLOv5, a cutting-edge deep learning model, for high-speed and accurate 2D object detection. The architecture of YOLOv5 is explored along with dataset preparation, preprocessing, training techniques, and comparative analysis with alternative models such as Faster R-CNN and SSD. Using the COCO and Pascal VOC datasets, YOLOv5 demonstrated significant performance with a mean Average Precision (mAP) of 0.85 and a frame rate of 35 FPS, achieving real-time inference. Challenges during implementation and future improvement directions are discussed to guide researchers and practitioners in deploying YOLO-based systems efficiently.