2019/07/10 by Tom Bruls, Bruls, Tom, Horia Porav +5
Computer Science · Engineering · Environmental Science · #Automated Road and Building Extraction #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Remote Sensing and LiDAR Applications #Robotics (cs.RO) #Video Surveillance and Tracking Methods
paper · pdf · doi:10.48550/arxiv.1907.04569
openalex publication_date 2019/07/10 · openalex created_date 2022/07/28 · openalex updated_date 2026/07/28
Road markings provide guidance to traffic participants and enforce safe\ndriving behaviour, understanding their semantic meaning is therefore paramount\nin (automated) driving. However, producing the vast quantities of road marking\nlabels required for training state-of-the-art deep networks is costly,\ntime-consuming, and simply infeasible for every domain and condition. In\naddition, training data retrieved from virtual worlds often lack the richness\nand complexity of the real world and consequently cannot be used directly. In\nthis paper, we provide an alternative approach in which new road marking\ntraining pairs are automatically generated. To this end, we apply principles of\ndomain randomization to the road layout and synthesize new images from altered\nsemantic labels. We demonstrate that training on these synthetic pairs improves\nmIoU of the segmentation of rare road marking classes during real-world\ndeployment in complex urban environments by more than 12 percentage points,\nwhile performance for other classes is retained. This framework can easily be\nscaled to all domains and conditions to generate large-scale road marking\ndatasets, while avoiding manual labelling effort.\n