Vision-based lane departure warning framework.

Area of Science:

• Computer Vision
• Automotive Engineering
• Road Safety

Background:

• Lane departure crashes are a significant cause of global road traffic accidents and fatalities.
• Driver error in judging vehicle path is a primary factor in these incidents.
• Existing automotive safety systems require enhancement for reliable lane departure detection.

Purpose of the Study:

• To propose and evaluate a vision-based lane departure warning framework.
• To enable accurate lane departure detection in both daytime and night-time driving conditions.
• To analyze performance across urban and highway road types in Malacca.

Main Methods:

• A two-stage lane detection process: pre-processing (color conversion, ROI extraction, segmentation) and Hough transform-based detection.
• Computation of a lateral offset ratio using detected lane boundary coordinates for warning.
• Real-life datasets encompassing diverse traffic, road, and lighting conditions for evaluation.

Main Results:

• The proposed framework achieved an average lane detection rate of 94.71%.
• An average lane departure detection rate of 81.18% was recorded.
• Performance was benchmarked against established methods and datasets like Caltech lanes.

Conclusions:

• The vision-based framework demonstrates satisfactory performance for lane departure warning.
• Challenges such as worn markings, low light, and occlusions impact false positive rates.
• Further research is needed to address these challenges for improved system robustness.