Fast Lane Detection Based on Improved Enet for Driverless Cars

Lane detection is an important module for autonomous vehicles and advanced driver assistance systems. Due to the limitation of vehicular environments' computing power, it is still a challenge to improve system speed and keep high accuracy simultaneously. In this paper, we propose a realtime method of lane detection based on improved Enet, which combines the advantages of channel shuffle and model compressing in model simplification. The comprehensive experiments demonstrate that our approach achieves a speed of 1.84 times (79.1FPS) of the original Enet (42.8FPS) on TuSimple dataset with a tradeoff between speed and accuracy. Furthermore, the size of our model is about four times smaller than the original model, and it is more suitable to use in the mobile onboard computing system.