Lane Detection with Centerline Calculation and Temporal Tracking for Autonomous Driving

Engineering notes

While deep learning has significantly advanced this field, existing methods often focus solely on lane segmentation without providing a continuous, navigable path. Trained and evaluated on the TuSimple benchmark dataset, our model achieves a competitive F1-Score of 93.37% and an accuracy of 93.06%.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

The precise detection of road lanes is a critical perception task for Advanced Driver-Assistance Systems (ADAS) and autonomous vehicles. While deep learning has significantly advanced this field, existing methods often focus solely on lane segmentation without providing a continuous, navigable path. This paper presents a complete end-to-end system that not only detects a variable number of lanes using instance segmentation but also introduces two major extensions: a robust module for calculating the geometric centerline of the current lane and a temporal memory system to ensure detection consistency across consecutive frames. These extensions are vital, as they transform raw pixel-level output into the smooth, actionable trajectory required by vehicle control systems. Our approach extends the work of Neven et al. by integrating a polynomial fitting and lane selection strategy to generate a smooth, actionable trajectory for vehicle control. The system is implemented as a modular pipeline with five integrated components: ingestion and preprocessing, neural network segmentation, polynomial fitting with centerline calculation, coordinate transformation, and temporal tracking. Trained and evaluated on the TuSimple benchmark dataset, our model achieves a competitive F1-Score of 93.37% and an accuracy of 93.06%. The result is a comprehensive and functional pipeline that provides an essential input for vehicle navigation, bridging the gap between raw perception and motion planning.

Links and sources

• Official / arXiv page

Comments