Monocular 3D lane detection for Autonomous Driving: Recent Achievements, Challenges, and Outlooks

TL;DR

This review summarizes and analyzes the current state of achievements in the field of 3D lane detection research, discusses the performance of these cutting-edge algorithms, analyzes the time complexity of various algorithms, and highlights the main achievements and limitations of ongoing research efforts.

Abstract

3D lane detection is essential in autonomous driving as it extracts structural and traffic information from the road in three-dimensional space, aiding self-driving cars in logical, safe, and comfortable path planning and motion control. Given the cost of sensors and the advantages of visual data in color information, 3D lane detection based on monocular vision is an important research direction in the realm of autonomous driving, increasingly gaining attention in both industry and academia. Regrettably, recent advancements in visual perception seem inadequate for the development of fully reliable 3D lane detection algorithms, which also hampers the progress of vision-based fully autonomous vehicles. We believe that there is still considerable room for improvement in 3D lane detection algorithms for autonomous vehicles using visual sensors, and significant enhancements are needed. This review looks back and analyzes the current state of achievements in the field of 3D lane detection research. It covers all current monocular-based 3D lane detection processes, discusses the performance of these cutting-edge algorithms, analyzes the time complexity of various algorithms, and highlights the main achievements and limitations of ongoing research efforts. The survey also includes a comprehensive discussion of available 3D lane detection datasets and the challenges that researchers face but have not yet resolved. Finally, our work outlines future research directions and invites researchers and practitioners to join this exciting field.

Figures (10)

• Figure 1: Chronological overview of 3D lane detection based on monocular images. Green arrows represent methods based on CNNs, orange arrows represent methods based on Transformers, and blue arrows represent hybrid architecture methods.
• Figure 2: An overview of the 3D-LaneNet garnett20193d architecture. Information is processed in two parallel streams or pathways: the image-view pathway and the top-view pathway. This is called the dual-pathway backbone.
• Figure 3: The procedures of Gen-LaneNet guo2020gen involve encoding an input image into deep features using a segmentation backbone, which are then decoded into a lane segmentation map. The 3D-GeoNet then focuses on geometry encoding and predicts intermediate 3D lane points represented in top-view 2D coordinates and real heights. Finally, a geometric transformation converts the network output into real-world 3D lane points.
• Figure 4: The architecture of Anchor3DLane huang2023anchor3dlane. Utilizing a front-view input image, we employ a CNN backbone and a Transformer layer to extract the visual feature $F$ initially. Subsequently, 3D anchors are projected to sample their features from $F$ based on camera parameters. Following this, a classification head and a regression head are utilized to make the final predictions. The lane predictions can also be used as new 3D anchors for iterative regression.
• Figure 5: The pipeline of PersFormer chen2022persformer. The key is to understand the spatial feature transformation from the front view to the BEV space, aiming to enhance the representativeness of the resulting BEV features at the target point by taking into account the local context around the reference point.