Proposing an Accurate Counting Method for Moving Objects by Optimizing Count-line Placement

Abstract

In recent years, traffic measurements using general object detection AI have been used in various scenes. In Japan, there is a movement to always use CCTV cameras installed on major roads to constantly observe traffic. In general, the traffic volume measurement AI uses a cross-sectional traffic volume measurement using the line separation judgment algorithm, but the criteria and judgment basis for setting this Count-line is not clearly shown. In this study, a comparative verification is performed using a Kernel Density Estimation to set up an appropriate Count-line and manually setting. As a result, changing the position of the Count-line resulted in a maximum difference of 95.7% in measurement accuracy.