The GAIT-IST dataset is made available to address de lack of publicly available pathological gait datasets, representing different types of gait, to be used in gait analysis and classification systems.
The database consists of sequences of images acquired from 10 subjects simulating pathological gait. Images are captured by a smartphone camera with 1280x720 pixel resolution, at the Multimedia signal Processing Group of Instituto de Telecomunicações – Instituto Superior Técnico, Lisbon, Portugal.
The IST-EURECOM LFFD database can be used not only in the context of face recognition research but also for other research areas such as emotion recognition, gender classification, age estimation, ethnicity classification and face modeling.
The types of gaits featured in the dataset, besides Normal gait, are Diplegic, Hemiplegic, Neuropathic and Parkinsonian. The dataset includes, for each gait type, sequences with 2 severity levels, 2 directions of walking and 2 repetitions per participant, except for normal gait, which does not have different severity representations. This results in a total of 10 x (4 x 2 x 2 x 2 + 2 x 2) = 360 gait sequences. Acquisitions considered the subject walking parallel to the camera plane, with a minimum of 2 complete gait cycles captured in each direction.
For each sequence, the results of a number of image pre-processing techniques are also present in the database. These include silhouette and skeleton images, and the Gait Energy image (GEI) and Skeleton Energy Image (SEI) representations. For each sequence, the complete gait cycles were cropped and used to compute the two representations. The silhouettes and skeletons included in the dataset are only the ones used to compute the GEIs and SEIs, i.e., the frames comprised in the detected complete gait cycles. All of these images were centered and normalized to a size of 224x224 pixels.
The files composing the database are organized according to a hierarchical structure. The root level of the hierarchy includes the folders for each of the five gait types (each stored in a separate zip file for download).
Each of these folders contains 10 subfolders, corresponding to each subject. In each subject's folder there are folders entitled: “GEIs”, “SEIs”, “silhouettes” and “skeletons”. Each of these, for the pathological gait sequences, contain subfolders named: “sub i pat j lvl k-r_direction”, where:
“i” is the number of the subject;
“j” is the number of the type of gait, which can be from 1 to 4 corresponding to Diplegic, Hemiplegic, Neuropathic and Parkinsonian, respectively;
“k” can be “1” or “2” corresponding to the two levels of severity, with “2” being the most severe;
“r” can also be “1” or “2”, identifying the two repetitions;
be “front” or “back”, indicating the walking direction of the sequence.
For the Normal sequences, these subfolders are
named: “sub i
the Normal sequences, these subfolders are named: “sub i normal-r_direction”.
· The database can be downloaded from the links below.
· A password for decrypting the compressed Zip files will be provided after receiving the duly signed license agreement (see above).
How to reference the database:
· J. Loureiro and P.L. Correia, “Using a Skeleton Gait Energy Image for Pathological Gait Classification”, Proceedings of the 15th IEEE International Conference on Automatic Face and Gesture Recognition (FG 2020), Buenos Aires, Argentina, 18 -22 May 2020
Feedback is welcome. Please provide your comments and suggestions by e-mail.