[IV-148] Empirical study on simplified traffic survey using image recognition system
Keywords:Traffic volume, Image recognition, Artificial intelligence, Deep learning, Annotation
There is a way to grasp the traffic volume by using the continuous observation equipment and images of the monitoring camera installed by the road manager.
However, most of them assume images shot from high places, and there are restrictions on the installation of shooting equipment.
In this study, we verified the shooting height that maximizes the accuracy by using videos of three different shooting heights using commercially available tripods that are easy to install.
As a result, it was found that the analysis using teacher data with a shooting height of 60 cm had the highest accuracy, and that the detection accuracy could be improved by using the teacher data as a mixed data set.
However, most of them assume images shot from high places, and there are restrictions on the installation of shooting equipment.
In this study, we verified the shooting height that maximizes the accuracy by using videos of three different shooting heights using commercially available tripods that are easy to install.
As a result, it was found that the analysis using teacher data with a shooting height of 60 cm had the highest accuracy, and that the detection accuracy could be improved by using the teacher data as a mixed data set.
Please log in with your participant account.
» Participant Log In