Lane line detection and departure estimation in a complex environment by using an asymmetric kernel convolution algorithm

Malik Haris,Jin Hou,Xiaomin Wang
DOI: https://doi.org/10.1007/s00371-021-02353-6
IF: 2.835
2022-01-10
The Visual Computer
Abstract:Deep learning has made tremendous advances in the domains of image segmentation and object classification. However, real-time lane line detection and departure estimates in complex traffic conditions have proven to be hard in autonomous driving research. Traditional lane line detection methods require manual parameter modification, but they have some limitations that are still susceptible to interference from obscuring objects, lighting changes, and pavement deterioration. The development of accurate lane line detection and departure estimate algorithms is still a challenge. This article investigated a convolutional neural network (CNN) for lane line detection and departure estimate in a complicated road environment. CNN includes a weight-sharing function that lowers the training parameters. CNN can learn and extract features frequently in image segmentation, object detection, classification, and other applications. The symmetric kernel convolution of classical CNN is upgraded to the structure of asymmetric kernel convolution (AK-CNN) based on lane line detection and departure estimation features. It reduces the CNN network's computational load and improves the speed of lane line detection and departure estimates. The experiment was carried out on the CULane dataset. The lane line detection results have high accuracy in a complex environment by 80.3%. The detection speed is 84.5 fps, which enables real-time lane line detection.
What problem does this paper attempt to address?