Robust Lane Detection via Expanded Self Attention

Abstract

The image-based lane detection algorithm is one of the key technologies in autonomous vehicles. Modern deep learning methods achieve high performance in lane detection, but it is still difficult to accurately detect lanes in challenging situations such as congested roads and extreme lighting conditions. To be robust on these challenging situations, it is important to extract global contextual information even from limited visual cues. In this paper, we propose a simple but powerful self-attention mechanism optimized for lane detection called the Expanded Self Attention (ESA) module. Inspired by the simple geometric structure of lanes, the proposed method predicts the confidence of a lane along the vertical and horizontal directions in an image. The prediction of the confidence enables estimating occluded locations by extracting global contextual information. ESA module can be easily implemented and applied to any encoder-decoder-based model without increasing the inference time. The performance of our method is evaluated on three popular lane detection benchmarks (TuSimple, CULane and BDD100K). We achieve state-of-the-art performance in CULane and BDD100K and distinct improvement on TuSimple dataset. The experimental results show that our approach is robust to occlusion and extreme lighting conditions.

Repainting and Imitating Learning for Lane Detection

“…The proposed RIL framework can be easily plug-and-play in most cutting-edge methods without any extra inference cost. Experimental results prove the eectiveness of RIL framework both on CULane [23] and TuSimple [33] for four modern lane detection methods including UFAST [26], ERFNet [27], ESA [11] and Cond-LaneNet [14] respectively.…”

“…In this section, we rst briey introduce the overview framework of our proposed RIL and then illustrate each module in RIL. Note that our method can be exibly combined with many cutting-edge lane detection methods such as [11,14,26,27]. In this section, we adapt ERFNet [27] to RIL framework for a demonstration.…”

“…Lane detection Recent approaches [4,16,19,25] focus on the deep neural networks and signicantly boost the lane detection performance due to the powerful representation learning ability. Some methods [11,23,27] treat lane detection as a semantic segmentation task. For instance, SCNN [23] designs slice-by-slice convolutions within feature maps to exchange pixel information between pixels across rows and columns in a layer.…”

Learning to Predict 3D Lane Shape and Camera Pose from a Single Image via Geometry Constraints

“…Deep methods built by convolutions have exploded in recent years, making significant progress and applicable systems for real applications (Chen et al 2018;Zhao, Yuan, and Chen 2020). Two-stage methods which extract segmentation or proposal plus post-processing ruled the filed for several years (Neven et al 2018;Philion 2019;Pan et al 2018;Zhang et al 2018;Hou et al 2019;Ko et al 2020;Li et al 2020;Tabelini et al 2021;Qin, Wang, and Li 2020;Xu et al 2020;Jung et al 2020;Yoo et al 2020;Lee et al 2021;Zheng et al 2020). To streamline the pipeline into an endto-end fashion, single-stage methods (Torres et al 2020;Liu et al 2021) directly estimate coefficients of prior mathematical curves have shown both higher efficacy and efficiency.…”

Lane-GAN: A Robust Lane Detection Network for Driver Assistance System in High Speed and Complex Road Conditions

“…With the evolution of advanced assisted driving system and automatic driving technology, the road accident probability is greatly reduced and the driving safety is improved [ 1 , 2 , 3 , 4 ]. As a key and challenging part of automatic driving and advanced assistant system, lane detection has also become a research hotspot [ 5 , 6 , 7 , 8 ]. It is vital for driver assistance systems to obtain the accurate location of each lane, which is also the goal of the lane detection algorithm.…”