Gait recognition by fusing direct cross-view matching scores for criminal investigation

被引：0

作者：

机构：

[1] Institute of Scientific and Industrial Research, Osaka University, Ibaraki

来源：

| 1600年 / Information Processing Society of Japan卷 / 05期

基金：

日本学术振兴会;

关键词：

Biometrics; Criminal investication; Cross-view; Fusion; Gait recognition;

D O I：

10.2197/ipsjtcva.5.35

中图分类号：

学科分类号：

摘要：

We focus on gait recognition for criminal investigation. In criminal investigation, person authentication is performed by comparing target data at the crime scene and multiple gait data with slightly different views from that of the target data. For this task, we propose fusion of direct cross-view matching. Cross-view matching generally produces worse result than those of same-view matching when view-variant features are used. However, the correlation between cross-view matching with different view pairs is low and it provides improved accuracy. Experimental results performed utilizing large-scale dataset under settings resembling actual criminal investigation cases, show that the proposed approach works well. © 2013 Information Processing Society of Japan.

引用

页码：35 / 39

页数：4

共 14 条

[11]

Muramatsu D., Shiraishi A., Makihara Y., Yagi Y., Arbitrary view transformation model for gait person authentication, Proc. 5th IEEE Int. Conf. Biometrics: Theory, Applications and Systems, pp. 1-6, (2012)

[12]

Nixon M.S., Tan T.N., Chellappa R., Human Identification Based on Gait, (2005)

[13]

Sarkar S., Phillips J., Liu Z., Vega I., Ther P.G., Bowyer K., The HumanID gait challenge problem: Data sets, performance, and analysis, IEEE Trans. Pattern Analysis and Machine Intelligence, 27, 2, pp. 162-177, (2005)

[14]

Sugiura K., Makihara Y., Yagi Y., Omnidirectional gait identification by tilt normalization and azimuth view transformation, Proc. IEEE Workshop on OMNIVIS, (2008)

← 1 2 →