[3Xin4] Poster session 1

[3Xin4-16] A Hybrid Person Identification Method Using an Image and Time Series Information for Security Cameras

〇Shuji Awai¹, Toshihiko Aoki¹, Shohei Ishizuka¹, Takeshi Konno¹ (1.Fujitsu Limited)

Keywords:gait recognition, ReID, security camera

Security cameras are being used to search for lost children and identify suspicious individuals. However, there is a problem that it takes time and effort to confirm visually from enormous images. Therefore, there is the research of image analysis by AI recently. As a conventional method, there is Person Re-identification (ReID) which extracts feature quantities by inputting a person image. However, there is a problem in which the accuracy lowers by change and occlusion of the lighting. In this paper, we propose a hybrid human identification method using images and time series. In the proposed method, features of ReID are extracted from a person image, and features of gait are extracted from attitude estimation information of time series. Then, the person collation is performed by using each similarity. In the experiment, it was confirmed that the proposed method improves the accuracy of the person collation in comparison with ReID.