Paper Title
YOLOv8 Adaptation for Real-Time Line Detection in Autonomous Navigation Systems for Robots

Abstract
Autonomous navigation in robotics requires robust and efficient visual perception systems capable of real-time environmental understanding. This paper presents a novel adaptation of the YOLOv8 object detection framework specifically optimized for line detection tasks in autonomous robot navigation systems. Traditional line detection methods often struggle with computational efficiency and accuracy trade-offs in dynamic environments, limiting their applicability in real-time autonomous systems. We adopt a strategy to alter the architecture of merely YOLOv8, yet adding particular convolutional layers and attention to the mechanism, helping in amplifying the linear feature extraction without loss of inherent specs of the framework such as speed. The modified model utilizes user-defined loss that aims to find fragmented lines and be orientation-specific with continuity, compensating typical topics in problematic domains line detection and angular resolution. We propose a slight modification to a lightweight feature pyramid network which performs 23 percent fewer computations than its counterparts and 15 percent better line detection accuracy than standard YOLOv8 versions. The system has been tested on a large data set that contained a variety of indoor and outdoor scenes such as corridors, roads, and structured pathways. The experiments confirm that our implementation of YOLOv8 is capable of performing real-time operations at 45-50 FPS using standard GPU processors with a 94.3% accuracy on line detection and 18% fewer false positives than standard methods of computer vision. This robustness and reliability of the system and the overall effectiveness of the system was verified in field tests of autonomous mobile robots in all possible situations. The modified YOLOv8 algorithm has successfully facilitated smooth path following, lane keeping and obstacle wiping actions in the dynamic conditions. The comparative analysis conducted with the current line detection computing algorithms such as modified versions of Hough Transform and CNN-based algorithms are proving to be a great success computationally and also in terms of the precision of detection. The work is relevant to the future of computer vision in robotics since it illustrates the effectiveness of implementing the latest object detection systems in the field of specialized perception used in robotics. The suggested variant of YOLOv8 provides a feasible offer in terms of real-time line detection of the autonomous navigation stack, and the scope may be widened to cover agricultural robots, warehouse automation, and autonomous vehicles. Keywords - Yolov8, Line Detection, Autonomous Navigation, Real-Time Processing, Computer Vision, Robotics