An Analysis on Bicycle and Left-Turn Vehicle Conflict with Trajectories Extracted from Faster Regions with Convolutional Neural Networks Method at Signalized Intersections

Abstract

As for the safety of bicycle traffic environment, it is desirable for bicycles to travel in the same direction as automobiles on the roadway because it is considered that the difficulty for car drives to recognize bicycles outside the roadway is a cause of accidents, and the development of a travel space based on traffic on the roadway is being promoted. In this study, video surveys were conducted at several signalized intersections with different intersection inflow conditions for bicycles and left-turning vehicles, and trajectory data from the faster Regions with Convolutional Neural Network Method (Faster R-CNN) was used to investigate the complex phenomenon of straight-line bicycles and left-turning vehicles. For the evaluation method, we focused on the change in speed during the time of conflict and used the quasi time-to-collision (TTC) and the speed difference through the intersection. As a result, it was found that while the speed of bicycles did not change much between different sections of the road, the presence or absence of complications and the position of the traffic were major factors in determining the speed.