Additional gait datasets for gait normality assessment

Introduction

Gait normality assessment is currently getting attention in biomechanics and computer vision research fields. However, most gait datasets only focus on the problem of action recognition and person identification. Due to the rarity of datasets for gait normality assessment, we adaptively generate two small datasets that are appropriate for simple normality-related tests. The details and download links of these two datasets are provided below.

Data description

The two datasets were created by adapting some mocap data sampled from CMU and SFU databases. The mocap data provide a collection of 3D joints forming 3D skeletons having the same structure as Microsoft Kinect 2. Besides, these mocap data were converted to point clouds by fitting a 3D model (created with MakeHuman) and using the set of 3D vertices as the point clouds. These Kinect skeletons and 3D point clouds are appropriate for researchers working on gait analysis with features extracted from  joint coordinates and/or point clouds.

Selected CMU gait sequences:

Selected SFU gait sequences:

Note: The lengths of these walking sequences are different and long sequences were split into multiple ones during the conversion.

Converted datasets:

Dataset Training set
(only normal walking gait)
Test set
Normal gait Abnormal gait
CMU 540 frames
(5 sequences)
769 frames
(8 sequences)
2224 frames
(7 sequences)
SFU 1082 frames
(5 sequences)
1295 frames
(6 sequences)
3086 frames
(13 sequences)

Download links of datasets:

Notice: the index of skeletal joints is the same as Kinect 2

Using the dataset

These two datasets can be used for research and educational purposes. If you employ them for your work, please adaptively add these texts to your acknowledgments section depending on which dataset is used:


Last update: March, 2019