LaCNet: Real-time End-to-End Arbitrary-shaped Lane and Curb Detection with Instance Segmentation Network

Accurate and robust detection of lane and curb in urban areas is essential to many real-world intelligent vehicle applications. Existing vision-based researches treat lane detection and curb detection separately due to the nature of curb detection problem relying on 3D features, which is not efficient when driving in a real world. This paper presents a unified network to incorporate these two tasks together by taking advantage of the powerful feature learning ability brought by deep convolutional neural networks. The resulting unified network provides valuable road boundary information by curb detection even when lane markings are not visible during vehicle navigation. Another significant capability coming with the proposed method is able to accurately differentiate various lane and curb instances with tiny gaps or complex spatial relationships which are thought as the biggest challenge in real driving situations. To achieve this, the network is specially designed to guide lane and curb instances to learnable kernel through pixel grouping. This learnable kernel is capable to handle any number of arbitrary-shaped lanes and curbs no matter which angle the vehicle is heading at. In the end, the presented approach is evaluated on two datasets (BDD100K and self-collected dataset) scoring 32 FPS processing speed. Results are very encouraging with more than 98% F₁ measure for both lane and curb detection on the self-collected dataset.