A two-stage real-time YOLOv2-based road marking detector with lightweight spatial transformation-invariant classification

作者：

Highlights：

• The proposed two-stage YOLOv2-based network tackles distorted road markings detection with balanced precision and recall.

• The embedded spatial transformer layer in the second stage conducts the detection to better precision.

• A new dataset is created for the public use with more than 11,000 high-resolution images under various conditions.

• The two-stage detection network achieves 86.5% mAP outperforming current real-time detection frameworks.

• The real-time network is able to run at 58 FPS in a single GTX 1070 under diverse circumstances.

摘要

•The proposed two-stage YOLOv2-based network tackles distorted road markings detection with balanced precision and recall.•The embedded spatial transformer layer in the second stage conducts the detection to better precision.•A new dataset is created for the public use with more than 11,000 high-resolution images under various conditions.•The two-stage detection network achieves 86.5% mAP outperforming current real-time detection frameworks.•The real-time network is able to run at 58 FPS in a single GTX 1070 under diverse circumstances.

论文关键词：Deep learning,Road marking,Spatial transform,Real-time object detection,Object classification

论文评审过程：Received 28 July 2019, Revised 4 May 2020, Available online 12 July 2020, Version of Record 22 July 2020.

论文官网地址：https://doi.org/10.1016/j.imavis.2020.103978