An Improved YOLOv5s for Lane Line Detection

Abstract

The lane line target detection is of great significance to the aspect of autonomous vehicles. At present, target detection needs lightweight models with high accuracy and real-time detection in the field of autonomous driving. The current more popular lane detection model takes up a lot of memory, which is difficult to deploy to mobile devices with small computing loads. Therefore, in order to settle the question of large memory footprint, this paper presents an improved YOLOv5s model that combines DWConv with GhostBottleneck to replace the CSP structure in YOLOv5s. Lane line detection is implemented by YOLOv5s and the improved YOLOv5s respectively. Experiments results indicate that the size of the improved YOLOv5s model proposed in this paper is reduced by three quarters and the detection speed is improved on the premise of sacrificing the micro accuracy(mAP@.5).