Advanced Lane Detection with Ensemble Models: A Hybrid Approach Leveraging ResNet and VGG Variants for Feature Extraction

Identifying lanes is an integral aspect of autonomous driving, which involves a road scene’s effective feature extraction. Therefore, in this paper, a performance comparison of the different CNN models targeting lane detection tasks including, but not exclusive to, ResNet-101, ResNet-50, VGG-16, and VGG-19 is presented. The ResNet family of models performs well due to the residual connections, enabling better feature-extracting capability at greater depths. At the same time, spatial detail capture is the strong point of the VGG family of networks, but at a higher cost in processing power. In addition, the potential of XGBoost to boost lane detection performance while using features extracted from a CNN is investigated. This analysis shows the performance and efficiency trade-offs that exist and assists in determining the best models that can be integrated into autonomous driving systems.