Edge-focused lightweight framework for real-time lane detection in driver assistance systems

This paper presents ELDA-Net, a lightweight deep architecture for real-time lane detection that balances accuracy, speed, and hardware efficiency for driver assistance systems. The model integrates a U-Net-based semantic segmentation backbone with a novel Adaptive Lane Estimation Mechanism (ALEM) to enhance prediction continuity in occluded and curved road scenarios without external dependencies. Evaluations demonstrate strong performance across benchmarks: 93.2% F1 score on TuSimple and 87.6% on the challenging CULane dataset, while maintaining 25 FPS on an NVIDIA Jetson Nano. The results confirm ELDA-Net as a practical, scalable solution for embedded lane detection in resource-constrained environments. The full implementation is available at: https://github.com/maijiddah/ELDA-Net.