Development of Traffic Measurement AI for Bicycles and Pedestrians,and study methods of utilizing the acquired information on pedestrian crossings

抄録

In recent years, image analysis methods using AI have been introduced to measure automobile traffic volume, and automatic observation systems have been constructed. On the other hand, the measurement accuracy of bicycles and pedestrians is still low, and expectations are growing for the development of traffic measurement AI that can obtain not only traffic volume but also movement speed and movement trajectory. In this study, we aim to develop traffic measurement AI that can estimate the traffic volume and movement speed of bicycles and pedestrians. Specifically, we trained a convolutional neural network (CNN) using new training labels, and it became possible to measure traffic volume and estimate speed using information obtained from object tracking. In addition, by visualizing the obtained movement speed and trajectory and combining it with TTC, which indicates the danger of collision between objects, it became possible to analyze and discuss each traffic mode in detail.