Object Detection Employing YOLOv8 in Conjunction with a Custom Dataset

Abstract

In a variety of applications, including surveillance and safety systems, object detection is a very important characteristic. The objective of this study is to apply the sophisticated object identification model known as YOLOv8 to a particular dataset that has been developed for emergency collision avoidance systems specifically. For the purpose of conforming to the format specifications of YOLOv8, the dataset was painstakingly hand-labeled and polished. It included components such as autos, pedestrians, and barriers in a variety of situations. In order to improve the efficiency of the model, supplementary data and transfer learning strategies were applied. According to the results of the experiments, the model achieves a mean average precision (mAP) of [0.43], which demonstrates both its accuracy and efficiency. The ability of the device to detect potential threats in real-time may make it possible for it to be included in complete safety systems. For the purpose of optimizing real-time processing, additional modifications will be investigated in subsequent research, and the concept will be applied to edge devices.