International Journal of Scientific Research and Engineering Development

Road traffic accidents continue to pose a significant threat to human safety, with human error as a primary contributing factor. This study presents a vision-based Advanced Driver Assistance System (ADAS) designed to improve road safety by utilizing cutting-edge computer vision and deep learning methods. The system integrates three vital components: lane detection, using CNN-based U-Net models for precise lane tracking; road segmentation, employing advanced image segmentation for a granular understanding of road scenes; and Forward Collision Warning (FCW), which proactively identifies and alerts drivers to potential collision risks. The proposed ADAS model is thoroughly tested using well-known datasets such as TuSimple and BDD100K, where it delivers impressive results, achieving an accuracy of 85.79%, a mean Intersection over Union (IoU) of 36.67%, a mean Average Precision (mAP) of 51.52%. Despite facing hardware constraints, the system maintains real-time responsiveness and dependable performance. By combining these features, the system effectively tackles key road-related issues, boosts driver awareness, and helps minimize human-induced errors, ultimately promoting safer driving conditions. This work adds valuable insight to the evolving field of AI-driven vehicle safety technologies and underscores the significant role of computer vision in preventing road accidents and creating more secure transportation systems