Vehicle lane markings segmentation and keypoint determination using deep convolutional neural networks

Abstract

Lane detection is used to detect the lane markings in a road scene between which the vehicle is driving and provide the accurate location and shape of each lane marking. It serves as one of the key techniques to enable modern, assisted, and autonomous driving systems. However, lane detection poses several challenges. The lane markings vary in their shapes, colors, and patterns. The lack of distinct features and the presence of several occlusions on the roads makes the use of conventional methods using handcrafted features less robust and computationally expensive. In this study, we propose a compact and efficient multi-stage Convolutional Neural Network (CNN) architecture which can learn both the lane markings segmentation and also the localization and shape of each lane in the form of key-points. The proposed model combines a lane mask proposal network with a lane key-point determination network to accurately predict the key-points that describe the left and right lane-markings of the vehicle lanes. The high running speed and low computational cost of the proposed method make it suitable for being deployed in the real world vehicle systems. Through simulation results, we also show that the proposed method is robust to a variety of weather conditions and highway driving scenarios.