Lane Detection for Autonomous Ground Vehicle using Residual U-Net with SAVE-Net [Safe Autonomous vehicle Network]

---

Abstract

This —The extraction of lane in a road from imagery has garnered significant attention in autonomous ground vehicle. In this study, we introduce semantic segmentation-based ANN that leverages the combined residual learning strength and the UNet architecture for effective lane of driving area extraction for Autonomous Ground Vehicle [AGV]. This network is builds and constructed by residual modules, following a same structure to UNet. This model offers two primary advantages: first, the inclusion of residual units simplifies the training of DNN; second, the dense skip like connections enhance information flow, enabling a more compact architecture with improved performances efficiently. We evaluate our network using a publicly available road dataset [CULane] and benchmark it against UNet and two additional latest deep learning methods for road lane extraction. The proposed method outperforms almost all compared approaches, highlighting its superiority over recent advancements in the field. The result of Residual U-Net combined with our Save Net [1] for the Safe Zone Driving Space to represent the final result.

---

Keywords

Lane Detection; Autonomous Ground Vehicle (AGV); Deep Neural Network (DNN); DNN based Residual UNet; SAVE-Net