PointLaneNet: Efficient end-to-end CNNs for Accurate Real-Time Lane Detection

Abstract

Lane detection is crucial for driving assistance systems such as lane keeping assist (LKA) and lane departure warning (LDW). In the conventional method, image processing operations such as edge detection and the Hough transform are utilized to recognize lane lines. Currently, the deep learning method for lane detection is dominated by semantic segmentation algorithms. Image pixels are classified by the deep neural network, and the lane line information is extracted by clustering and other post-processing methods. However, these methods have problems such as large models and unintuitive training. This paper proposes an end-to-end lane detection algorithm that can obtain the coordinates of lane line points directly from the drive scene, and can be deployed on the embedded platform NVIDIA PX2. The algorithm satisfies the requirements of high precision and real-time performance in autonomous driving.