Gait Recognition Using 2D Poses
Over the last decades, the field of biometrics has become an important ally for human identification, mainly used for fraud prevention and access control in restricted areas, with the final purpose of increasing the security of the individuals in society. Nowadays,the most common biometric systems a...
| Authors: | , |
|---|---|
| Format: | article |
| Status: | Published version |
| Publication Date: | 2021 |
| Country: | Brasil |
| Institution: | Sociedade Brasileira de Computação (SBC) |
| Repository: | Revista Eletrônica de Iniciação Científica |
| Language: | English |
| OAI Identifier: | oai:journals-sol.sbc.org.br:article/2081 |
| Online Access: | https://journals-sol.sbc.org.br/index.php/reic/article/view/2081 |
| Access Level: | Open access |
| Keyword: | Biometrics Gait Recognition Pose Estimation |
| Summary: | Over the last decades, the field of biometrics has become an important ally for human identification, mainly used for fraud prevention and access control in restricted areas, with the final purpose of increasing the security of the individuals in society. Nowadays,the most common biometric systems are those based in features like fingerprints, face and iris. Despite the great performance of state-of-art methods that use these traits, an important challenge remains, which is the automatic human identification in low-resolution videos, at a distance and without the need for subject cooperation. In this context, the usual biometric systems do not meet the expected performance, and using gait features to identify individuals may be the only viable option. The goal of this work is to propose a new method for gait recognition using gait information extracted from 2D poses estimated over video sequences with high accuracy and low computational cost when compared to other state-of-art methods. In order to estimate the 2D poses, we use OpenPose, an open-source and robust pose estimator. The proposed new method was assessed in two public gait datasets, CASIA Gait Dataset-A and CASIA Gait Dataset-B, and obtained recognition rates comparable with state-of-the-art results, but using smaller feature vectors. |
|---|