# Keep your Eyes on the Lane: Real-time Attention-guided Lane Detection

**Authors:** Lucas Tabelini, Rodrigo Berriel, Thiago M. Paixão, Claudine Badue, Alberto F. De Souza, Thiago Oliveira-Santos  
**arXiv:** [2010.12035](https://arxiv.org/abs/2010.12035)  
**Date:** 2020-10-22  
**Summary page:** https://www.summarizepaper.com/en/paper/2010.12035v2/

## Key points

- Lane detection methods have made significant progress in complex real-world scenarios
- Many of these methods struggle to maintain real-time efficiency, which is crucial for autonomous vehicles
- The proposed LaneATT model uses anchors for the feature pooling step, similar to other generic deep object detectors
- The researchers developed a novel anchor-based attention mechanism that aggregates global information from the entire image to accurately infer lane positions in challenging conditions such as occlusion or missing lane markers
- The LaneATT model outperformed current state-of-the-art methods in terms of both efficacy and efficiency on three widely used datasets in the literature
- An ablation study was conducted and various efficiency tradeoff options were discussed that could be useful in practice
- Overall, the proposed LaneATT model offers a promising solution for real-time lane detection with high accuracy and efficiency.

## Summary

In recent years, lane detection methods have made significant progress in complex real-world scenarios. However, many of these methods struggle to maintain real-time efficiency, which is crucial for autonomous vehicles. To address this issue, a team of researchers led by Lucas Tabelini proposed LaneATT: an anchor-based deep lane detection model that uses anchors for the feature pooling step, similar to other generic deep object detectors. Since lanes follow a regular pattern and are highly correlated, the researchers hypothesized that global information may be essential in some cases to infer their positions accurately, particularly in challenging conditions such as occlusion or missing lane markers. Thus, they developed a novel anchor-based attention mechanism that aggregates global information from the entire image. The LaneATT model was extensively evaluated on three widely used datasets in the literature and outperformed current state-of-the-art methods in terms of both efficacy and efficiency. The researchers also conducted an ablation study and discussed various efficiency tradeoff options that could be useful in practice. Overall, the proposed LaneATT model offers a promising solution for real-time lane detection with high accuracy and efficiency. This research has significant implications for autonomous vehicle technology and could contribute to safer driving experiences on our roads.