Intersection and Stop Bar Position Extraction from Crowdsourced GPS Trajectories

1 Detailed road features (e.g., lane marks and stop bars) are crucial for many recent intelligent 2 transportation system applications, especially for automated or autonomous driving systems. In 3 this paper, a crowdsourcing based method is proposed to mark intersection areas and map stop bar 4 positions without prior knowledge of road information. The proposed method includes an efficient 5 approach for marking intersection areas by analyzing the entropy of moving direction, as well as 6 a statistical model of stop positions for estimating the number and coordinates of stop bars. The 7 proposed method is applied to the real-world dataset collected for the Safety Pilot Model 8 Deployment Program (SPMDP). The numerical analysis results prove its applicability and 9 robustness in processing GPS trajectories of an urban region (a 1.2 km by 2 km rectangular area). 10 For the intersections covered well by trajectories, the accuracy of marking intersections is 95.7%. 11 For stop bar positioning, the mean and standard deviation of the errors are 0.25 m and 0.32 m. 12 13