Deep Learning Based Segmentation Approach for Automatic Lane Detection in Autonomous Vehicle

Engineering notes

Key topics: autonomous driving, autonomous vehicle, lane detection, semantic segmentation. See the paper for implementation details and experimental results.

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

Lane detection is an essential feature of autonomous vehicle systems, and it can be done automatically. In this research, a Deep Learning (DL)-based segmentation strategy for reliable and rapid lane detection is suggested. In order to study lane identification in autonomous driving scenarios, data will be gathered from the tuSimple website, which offers annotated images created for that purpose. Pre-processing the obtained data entails actions like resizing, normalising, and enhancing to guarantee consistent and high-quality inputs for the models. In this study, the widely-used DL models SegNet and U-Net are implemented to segment lanes. In order to facilitate the classification of lanes and non-lane regions at the pixel level, these models were developed with semantic segmentation tasks in consideration. Mean squared error (MSE), average miss, and computation time are some of the measures that use to assess the effectiveness of both models. The findings prove that both DL-based methods successfully locate lanes. The optimal model for lane recognition in autonomous vehicles are determined based on the evaluation metrics. The best model is the one with the smallest MSE and the smallest average miss value. For real-time applications, computation speed is also critical, therefore the study prioritize models that strike a good compromise between the two.

Links and sources

• Official / arXiv page

Comments