Detecting and Tracking of Road Boundary Lines Using Machine Learning Approach

Objectives: The study aims to improve road safety by developing a lane detection model that achieves high accuracy even in adverse conditions like poor lighting, occlusions, and inclement weather. In high-traffic countries like India, where poor lane discipline contributes to road accidents, this system can significantly enhance driver assistance. Methods: This study compares three approaches—Canny Edge Detection, Hough Transform, and a CNN-based model. The CNN was trained on a dataset of over 1,000 annotated images featuring diverse scenarios, including curves, shadows, and varied lighting conditions. Evaluation metrics include precision, recall, F1-score, and computational latency. Findings: The CNN-based model achieved an accuracy of 96%, and F1-score of 0.92, outperforming traditional methods, which showed an average accuracy of 70% and F1-score of 68%. It maintained robustness under low-light conditions (accuracy: 91%) and adverse weather (F1-score: 89%). Real-time frame processing improved to 25 FPS, marking a 30% enhancement over baseline. Novelty: By leveraging CNNs for feature extraction and training on a diverse dataset, this study demonstrates a significant improvement over traditional methods. The proposed approach bridges the gap between accuracy and real-time application, offering a viable solution for lane detection in challenging environments. Keywords: Accidents, Computer Vision, ML, CNN, Lane-Detection