Detection of Dangerous Objects By Pan-tilt Camera Yurika Fujii†, Shin-ichi Ito†, Momoyo Ito‡, Minoru Fukumi‡ Tokushima University 2-1, Minami-Josanjima, Tokushima, 770-8506, Japan †{c501837044, s.ito}@tokushima-u.ac.jp ‡{momoito, fukumi}@is.tokushima-u.ac.jp ABSTRACT Security cameras are increasing in public facili- ties. The main reason is improvement of security. Furthermore, security cameras help quick arrest of criminals. It is, therefore, important that we install security cameras. However, we expect to obtain too much data of movie by increasing the number of cameras too. Security guard cannot watch all movie of all cam- eras always. For these reasons, we thought not only getting data but also security improvement. If humans have dan- gerous objects, the movie’s situation should define them as dangerous. First, we detect humans in cam- era’s images. Second, if we detect dangerous ob- jects in human detection area, we think that there are dangerous humans. We use SSD to detect hu- mans and dangerous objects. KEYWORDS SSD(Single Shot MultiBox Detector), Security Camera, Security, Image Processing, Deep learn- ing 1 INTRODUCTION Recently, security cameras are increasing in public facilities. The main reason is im- provement of security. Furthermore, secu- rity cameras help quick arresting criminals. This graph(Figure 1) shows the number of crimes decreased in the area installed security cameras[1]. Although security cameras are important for crime prevention, security guard cannot man- age them if the number of cameras are in- creased too much. For these reasons, we Figure 1. Number of crimes in the area installed security camera. The place is 6 regions in Tokyo. thought not only getting data but also improve- ment of security. However, we expect to ob- tain too much data of movies by increasing the number of cameras too. Security guard cannot watch all movies of all cameras always. Accordingly, we thought security cameras as- sisted too much data’s management and im- proved security. It is difficult to identify a dangerous person from motion video. For ex- ample, we don’t know why humans are run- ning outside such as training or chasing peo- ple. Their situations depend on the their pur- pose. Because, we judged by whether people have dangerous objects. We judge it as danger- ous situation when we detect dangerous objects and people at the same time. As a pre-stage of the video, we detect dan- gerous objects and humans from an image. It is important that cameras are low-cost, because purpose of this paper is detection of dangerous objects by security cameras. ’Dangerous object’ in this paper is Kitchen knife. Because, a knife is close to us and it has high killing power. Proceedings of The Fifth International Conference on Electronics and Software Science ICESS2019, Japan, 2019 ISBN: 978-1-941968-58-1 ©2019 SDIWC 61