A HIGH-PERFORMANCE DEEP LEARNING BIFPN–YOLOV8 MODEL FOR REAL-TIME AUTONOMOUS DRIVING APPLICATIONS

Abstract

Artificial intelligence (AI) and advanced sensor technologies have suggestively improved intelligent transportation systems and self-driving cars. Perception, a vital component, captures real-time traffic data important for many system functions, including agent behavior prediction. Accurate and timely object detection is an significant element of safe and reliable self-driving systems. Traditional detection frameworks frequently struggle with multi-scale object representation, resulting in decreased accuracy when recognizing small, distant, or veiled objects in complex traffic settings To overcome this issue, we present an improved object detection model that combines the Bi-Directional Feature Pyramid Network (BiFPN) and YOLOv8. The BiFPN allows efficient multi-scale feature fusion through weighted bidirectional connections, ensuring that both low-level fine details and high-level semantic cues are efficiently retained. This improved feature representation, combined with the lightweight and fast YOLOv8 detection head, permits the model to whole greater accuracy while maintaining real-time performance. Experimental results on benchmark autonomous driving datasets found important improvements in precision, recall, and mean Average Precision (mAP) related to baseline YOLOv8, mainly for small and partially obscured objects such as pedestrians, bicycles, and traffic signs. The recommended framework provides a robust and scalable solution for intelligent vehicle perception, paving the way for more secure and efficient self-driving applications.