Lane detection plays a vital part in autonomous driving. Conventional studies rely on less robust hand-craft features, while deep learning has improved the performance of lane detection to a great extent. Different from dominant methods based on semantic segmentation, this paper proposes an end-to-end framework named DevNet, which combines deviation awareness with semantic features based on point estimation. It consists of two modules to capture more representative features by integrating information of distance deviation and angle which helps to tackle diverse driving conditions in real environments, such as dim or shiny light conditions, crowdedness, and vanishing lanes. Experiments on public datasets indicate that the proposed method achieves favorable performance when compared with the state-of-the-art methods.
DevNet: Deviation Aware Network for Lane Detection
IEEE Transactions on Intelligent Transportation Systems ; 23 , 10 ; 17584-17593
2022-10-01
4939728 byte
Article (Journal)
Electronic Resource
English
LANE DEVIATION SUPPRESSION APPARATUS AND LANE DEVIATION SUPPRESSION METHOD
European Patent Office | 2018
|LANE DEVIATION SUPPRESSION DEVICE AND LANE DEVIATION SUPPRESSION METHOD
European Patent Office | 2018
|LANE DEVIATION MODE DETERMINATION DEVICE AND LANE DEVIATION ALARM DEVICE
European Patent Office | 2018
|TRAFFIC LANE DEVIATION WARNING DEVICE AND TRAFFIC LANE DEVIATION WARNING METHOD
European Patent Office | 2017
|