We address the problem of gait-based people identification by using two main approaches: from 2D information, and from 3D information. Moreover, we have created a new multi-view gait database which contains gait sequences of 20 actors that depict ten different trajectories each, including curved paths.

Gait recognition from 2D information.

From sequences of RGB data, we extract short-term dense trajectories that are described with the Divergence-Curl-Shear (DCS) descriptor. Then, the Pyramidal Fisher Motion (PFM) descriptor is computed to obtain a video-level gait descriptor.

The following publications present our results by using as basis the previously explained approach:



M. Marin-Jimenez, F.M. Castro, A. Carmona-Poyato, N. Guil
On how to improve tracklet-based gait recognition systems
Pattern Recognition Letters (PRL)
Project | Code

F.M. Castro, M. Marin-Jimenez, R. Medina-Carnicer
Pyramidal Fisher Motion for Multiview Gait Recognition
International Conference on Pattern Recognition (ICPR)
Video Code | Dataset

Gait recognition from 3D information.

These approaches require multiple camera viewpoints to obtain 3D information of the target person. Then, different descriptors are proposed in the following publications:

D. López-Fernández, F. J. Madrid-Cuevas, A. Carmona-Poyato, R. Muñoz-Salinas, R. Medina-Carnicer
A new approach for multi-view gait recognition on unconstrained paths
Journal of Visual Communication and Image Representation (JVCI)

D. López-Fernández, F. J. Madrid-Cuevas, A. Carmona-Poyato, M.J.
Marín-Jiménez, R. Muñoz-Salinas, R. Medina-Carnicer

Viewpoint-Independent Gait Recognition through Morphological Descriptions of 3D Human Reconstructions
Image and Vision Computing (IMAVIS)

D. López-FernándezF. J. Madrid-CuevasA. Carmona-PoyatoR. Muñoz-SalinasR. Medina-Carnicer
Entropy volumes for viewpoint-independent gait recognition
Machine Vision and Applications (MVAP)

D. López-FernándezF. J. Madrid-CuevasA. Carmona-PoyatoR. Muñoz-SalinasR. Medina-Carnicer
Multi-view gait recognition on curved trajectories
9th International Conference on Distributed Smart Camera (ICDSC 2015)

The AVA Multi-View Gait Dataset (AVAMVG)

To test some of our approaches, we have created a new multi-view gait database which contains gait sequences of 20 actors that depict ten different trajectories each, including curved paths. This database has been specifically designed to test Gait Recognition algorithms based on 3D data. Thus, the cameras have been calibrated and methods based on 3D reconstructions can use this dataset to test. Binary silhouettes are also provided.

Read more.

D. López-FernándezF. J. Madrid-CuevasA. Carmona-Poyato, M.J Marín-Jiménez, R. Muñoz-Salinas
The AVA Multi-View Dataset for Gait Recognition
2nd Workshop Activity Monitoring by Multiple Distributed Sensing (AMMDS 2014).