International Conference on Computer and Knowledge Engineering

Home / 12th International Conference on Computer and Knowledge Engineering

LPCNet: Lane detection by lane points correction network in challenging environments based on deep learning

Authors :

Sina BaniasadAzad¹ Seyed Mohammadreza Mousavi mirkolaei²

1- School of Electrical Engineering, Iran University of Science and Technology 2- School of Electrical Engineering, Iran University of Science and Technology

Keywords :

Lane Detection،Deep Learning،lane segmentation،unstructured road،Autonomous vehicles

Abstract :

Recently, lane detection methods with the help of deep learning have achieved significant accuracy in various conditions. But many do not perform well in computational complexity and are not applicable for real applications. In addition, their accuracy decreases in challenging situations. In this article, after identifying the critical points by the hourglass algorithm, we remove the redundant and invalid points using the Random Sample Consensus (RANSAC) algorithm. The line point correction network (LPCNet) achieves acceptable accuracy in challenging conditions such as desert roads with poor texture and foggy conditions with poor light and clarity. Also, the computational complexity of the system is suitable, making it practical for real-time execution. The number of network parameters is 4.5 million and significantly reduced compared to the valid methods. The execution speed in the tensor version reaches 32 frames per second, and the accuracy of the network in unlined environments is estimated at 49.80%.