BezierFormer: A Unified Architecture for 2D and 3D Lane Detection

Zhiwei Dong,Xi Zhu,Xiya Cao,Ran Ding,Wei Li,Caifa Zhou,Yongliang Wang,Qiangbo Liu

2024-04-25

Abstract:Lane detection has made significant progress in recent years, but there is not a unified architecture for its two sub-tasks: 2D lane detection and 3D lane detection. To fill this gap, we introduce BézierFormer, a unified 2D and 3D lane detection architecture based on Bézier curve lane representation. BézierFormer formulate queries as Bézier control points and incorporate a novel Bézier curve attention mechanism. This attention mechanism enables comprehensive and accurate feature extraction for slender lane curves via sampling and fusing multiple reference points on each curve. In addition, we propose a novel Chamfer IoU-based loss which is more suitable for the Bézier control points regression. The state-of-the-art performance of BézierFormer on widely-used 2D and 3D lane detection benchmarks verifies its effectiveness and suggests the worthiness of further exploration.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is: **Construct a unified architecture for simultaneously handling 2D and 3D lane detection tasks**. Specifically, although current lane detection methods have made significant progress in 2D and 3D lane detection, there is a lack of a unified architecture that can handle these two subtasks simultaneously. This results in a great deal of effort being required to adapt the state - of - the - art 2D lane detection model to 3D lane detection (or vice versa), and hinders the further development of lane detection technology.

To solve this problem, the author introduced **BézierFormer**, a unified 2D and 3D lane detection architecture based on Bézier curve representation. Bézier curves can efficiently represent 2D and 3D curves through a small number of control points, and are therefore suitable for unified lane representation. The main features of BézierFormer include:

1. **Bézier curve representation**: Use Bézier curves to represent lane curves, and efficiently represent 2D and 3D lanes through a small number of control points.
2. **Bézier curve attention mechanism**: Propose a new Bézier curve attention mechanism, which realizes comprehensive and accurate feature extraction of slender lane curves by sampling and fusing multiple reference points on each curve.
3. **Chamfer IoU loss function**: Propose an IoU loss function based on Chamfer distance, which is more suitable for Bézier control point regression and helps to learn the overall shape of the target curve more effectively.

Through these innovations, BézierFormer has achieved state - of - the - art performance in widely - used 2D and 3D lane detection benchmark tests, verifying its effectiveness and the value of future exploration.