Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces a lightweight highway lane detection architecture using 3D-ResNet and PINet to capture spatiotemporal information. Two models are proposed: one with FPN and self-attention, and another with a ROI detection head to focus on lane-relevant regions. The ROI-aware model achieves 93.40% accuracy on the TuSimple dataset with fewer parameters and reduced latency compared to existing methods, making it suitable for ADAS.
Achieve state-of-the-art lane detection accuracy with a model that's lightweight enough for real-time ADAS deployment.
This paper presents a lightweight, end-to-end highway lane detection architecture that jointly captures spatial and temporal information for robust performance in real-world driving scenarios. Building on the strengths of 3D convolutional neural networks and instance segmentation, we propose two models that integrate a 3D-ResNet encoder with a Point Instance Network (PINet) decoder. The first model enhances multi-scale feature representation using a Feature Pyramid Network (FPN) and Self-Attention mechanism to refine spatial dependencies. The second model introduces a Region of Interest (ROI) detection head to selectively focus on lane-relevant regions, thereby improving precision and reducing computational complexity. Experiments conducted on the TuSimple dataset (highway driving scenarios) demonstrate that the proposed second model achieves 93.40% accuracy while significantly reducing false negatives. Compared to existing 2D and 3D baselines, our approach achieves improved performance with fewer parameters and reduced latency. The architecture has been validated through offline training and real-time inference in the Autonomous Systems Laboratory at City, St George's University of London. These results suggest that the proposed models are well-suited for integration into Advanced Driver Assistance Systems (ADAS), with potential scalability toward full Lane Assist Systems (LAS).