Lane-Level Road Network Construction Based on Street-View Images

With the advancement of autonomous driving technologies, road network data have attracted a lot of attention as a virtual source of information. Traditional node–arc road networks are no longer able to match the demands of high-precision location awareness. Thus, lane-level road networks with more information have become a research hotspot. Furthermore, street-view pictures are a popular data source for building a lane-level road network because they provide a significant quantity of road information. The current method of constructing lane-level road networks based on street-view images performs feature extraction in image space and then projects it into geographic space. Hence, due to perspective and other rules, there are conflicts and overlaps in the exact locations after projecting the results of the street-view pictures from different viewpoints into geographic space. Because, to the best of the author's knowledge, there is no process for optimizing the overall geographic space results, the current study does not meet the demand for the accurate and comprehensive acquisition of lane-level road network in complete areas. However, this study proposed a lane-level road network construction method based on street-view image data, focusing on aggregating and optimizing the picture space extraction results in geographic space to improve the accuracy while aligning the results more consistent with the vector data requirements of geographic information systems. The experimental results show that by using street-view picture data, this technology can establish a submeter lane-level road network, which can be used for low-cost road data collection and updating.