Summary
International Workshop on Smart Info-Media Systems in Asia
2021
Session Number:RS1
Session:
Number:RS1-6
Estimation of Personal Information Based on Joint Locations Obtained by Pose Estimator
Shiho Hanashiro, Yui Suzuki, Ryusuke Miyamoto,
pp.46-51
Publication Date:2021/9/20
Online ISSN:2188-5079
DOI:10.34385/proc.66.RS1-6
PDF download (1.6MB)
Summary:
Estimating personal information using image information is important for analyzing actual social activities. The most popular visual cue for this purpose is facial imaging, however, it is difficult to achieve a high enough accuracy for practical use when the resolution of the facial images is not very high, and the image resolution becomes lower quite easily if the view point is far from the target. Gait may be a powerful cue to obtain accurate classification results about personal identification and gender prediction, using images captured by surveillance cameras; details of human parts are not required for gait analysis. Previous research has shown that the movement of joint locations may be a good feature for gait analysis, but joint locations cannot be estimated directly from input images. In this paper, a novel scheme for gait analysis using the movement of joint locations estimated using only input images is proposed. The experimental results showed that the proposed scheme achieved accurate results when Alpha Pose was used to obtain joint locations with a 93.13 % accuracy rate for personal identification and 93.56 % for gender prediction.