Improved and Efficient Object Detection Algorithm based on YOLOv5

Abstract

Object detection is a fundamental and impactful area of exploration in computer vision and video processing, with wide-ranging applications across diverse domains. The advent of the You Only Look Once (YOLO) paradigm has revolutionized real-time object identification, particularly with the introduction of the YOLOv5 architecture. Specifically designed for efficient object detection, YOLOv5 has enhanced flexibility and computational efficiency. This study systematically investigates the application of YOLOv5 in object identification, offering a comprehensive analysis of its implementation. The current study critically evaluates the architectural improvements and additional functionalities of YOLOv5 compared to its previous versions, aiming to highlight its unique advantages. Additionally, it comprehensively evaluates the training process, transfer learning techniques, and other factors, advocating the integration of these features to significantly enhance YOLOv5's detection capabilities. According to the results of this study, YOLOv5 is deemed an indispensable technique in computer vision, playing a key role in achieving accurate object recognition. The experimental data showed that YOLOv5-tiny performed better than anticipated, with a mean Average Precision (mAP) of 60.9% when evaluated using an Intersection Over Union (IoU) criterion of 0.5. Compared to other approaches, the proposed framework is distinguished by significant improvements in the mean average accuracy, computational flexibility, and dependability. As a result, YOLOv5 is suitable for a wide range of real-world applications, since it is both sophisticated and resilient in addressing present issues in the fields of computer vision and video processing.