YOLOv8 Based Object Detection for Self-driving Cars
Zakia Afrin, Fariya Tabassum, Hafsa Binte Kibria, MD. Rafi Imam, Md. Rokibul Hasan
Abstract
The investigation focuses on improving object detection in roadmaps for self-driving cars using the YOLOv8 model. Object detection is a pivotal component in developing autonomous vehicles, enabling them to identify and track diverse objects like cars, trucks, pedestrians, bicyclists, and traffic lights in their natural surroundings. The central goal of this study is to elevate the accuracy and efficiency of object detection methods for self-driving cars. The critical methodology employed in this research is the "You Only Look Once" (YOLO) approach. YOLO is indispensable for real-time object detection, allowing the vehicle to process a single image in a single forward pass, facilitating rapid decision-making. Object detection holds paramount importance in the realm of self-driving cars as it empowers vehicles to perceive and respond to their environment effectively. YOLO addresses this imperative by introducing a swift and precise object detection algorithm. This study’s hardware configuration meticulously designed for computational tasks ensures efficient processing, a critical aspect for real-time applications in self-driving cars. The integration of advanced hardware accelerators significantly contributes to the study’s success. The model’s mean Average Precision (mAP) achieved an impressive 77.8%, underscoring its proficiency in accurately detecting objects across various scenarios. The YOLOv8 model met and surpassed expectations through this research, providing robust object detection capabilities in real-world situations. This remarkable outcome emphasizes the YOLOv8 model’s effectiveness in self-driving cars, offering dependable object detection for safe navigation.