Camera Calibration Using Street Surveillance Camera for Detecting Wandering Elderly

Natsuki Miyahara, Takahiro Yoshioka, Takeshi Konno

Abstract: In this study, we propose a camera calibration method intended for detecting wandering elderlies using surveillance cameras. Using camera calibration, it is possible to obtain various characteristics of the elderly automatically, such as physical information and the size of belongings. Because the previous method has only shown its effectiveness for different situations from surveillance cameras, we propose a method to achieve high reliability for camera calibration towards dataset of possible shooting conditions in surveillance cameras. Furthermore, in the previous method, it was assumed that multiple people were shown in the video, which contributed to the construction of the geometric relationship in the estimation process. Hence, it is difficult to achieve high reliability of camera calibration for a video showing only one person, although wandering incident could occur without anyone around the elderly. Therefore, we propose a method for increasing constraints in the construction of geometric relationships by assigning multiple IDs instead of assigning one ID to the person. As a result, our method achieved a minimum camera height estimation error of 3.2 cm. We will continue this research for a more reliable camera calibration, and we further aim to construct a detection system for wandering elderlies to contribute to a safe and secure society.