Traffic Mirror Detection and Annotation Methods from Street Images of Open Data for Preventing Accidents at Intersections by Alert

Da Li, Hikaru Hagura, Taichi Miyabashira, Yukiko Kawai, Shintaro Ono; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2023, pp. 3264-3270

In recent years, research on autonomous driving has been actively pursued in the automotive industry. In Japan, a bill to revise the Road Traffic Act law regarding level 4 autonomous driving is passed in 2022, indicating a proactive approach toward autonomous driving. In light of these trends, improving safety during car travel has become an even more important challenge than before. Especially, intersections with poor visibility have been one of the major causes of traffic accidents, and improving safety at such intersections is an essential element in enhancing safety during car/bicycle travel. In this study, we aim to develop a system capable of identifying blind spots reflected in traffic mirrors by analyzing worldwide open data such as road images (e.g., Google Street View) and road information (e.g., OpenStreetMap). By Annotating the critical points from open data, the application could provide alerts to pedestrians and vehicles, enhancing safety in the vicinity of these blind spots. Specifically, we initially investigate the most effective deep learning model for detecting traffic mirrors. Additionally, we analyze the location information of traffic mirrors from geospatial data and road image data to construct a traffic mirror distribution map. Furthermore, we intend to equip bicycles with smartphones to track and detect the trajectories of pedestrians and vehicles reflected in these traffic mirrors.