The WeightGait Dataset
A multi-factorial dataset for simulated gait abnormality detection and assessment
Introduction
We introduce the WeightGait dataset - a dataset developed for facilitating vision-based gait assessment methodologies with more realistic conditions comparable to real world use.
The motivation for this dataset is to create a testing environment for gait assessment algorithms that is closer to the realities of application. To accomplish this, unlike other similar datasets, we do two main things uniquely:
- We simulate overlapping abnormalities, for a total of 9 different combinations of abnormality detailed below.
- The background and equipment used are imperfect and noisy to simulate the similar hardship experienced when trying to install a gait monitor into someone's home. This means cheap recording equipment for scalability resulting in relatively low-frames per recording. It also means slight feet/head clipping at times, only a single camera view to detect depth and no curation to the background or the clothing/walking speed of the participants.
The original 2D joint positions are estimated on the original videos using a lightweight implementation of the algorithm given in the paper 'HigherHRNet'. The joints in this dataset are given in their raw pixel format [column, row, depth]. Instructions and best practices for converting these to centimeters are provided in the ReadMe.md accompanying the dataset. The latest version of the Readme.md file is here.
The dataset was recorded using a single Intel Realsense D415 RGB-D camera with an average framerate of 28.5 fps.
Dataset Information
Characteristics
- 32 Participants, each with 60 short video clips. Each clip is cut down to approximately 60 RGB and 60 depth frames after empty or noisy frames are removed.
- Each participant is recorded for 20 clips of each of the following 3 conditions, mimicked using attachable 2.5 Kg leg weights: normal walk (no weight), limping (weight on one leg), shuffling (weights on both legs).
- Within each set of 20 clips, there are 3 further confounding factors: normal walking, walking and exhibiting a gait freeze event, walking over a small obstacle.
- All participants are medically "healthy", in the sense that no participant had a diagnosed ailment that is known to affect normal gait for their age.
Demographics
- The median participant age range is 25-29 with the youngest being 17 and the oldest 69 years old.
- 12 nationalities are featured in the dataset, across Asia, North America, Northern and Southern Europe, the middle east and West Africa.
- There are 19 male and 13 female participants, representing an approximate 60/40 split.
Sample Data
These are the sample files (links in the 'downloadables' column) of how the data for each individual subject looks like. Please note that the sample poses are the True poses of the subjects.
| Activity | Instance Code | Weight | Freeze | Obstacle | Downloadables |
|---|---|---|---|---|---|
| Regular Walk | 0 | None | No | No | [video] |
| Regular Walk w/ Gait Freeze | 1 | None | Yes | No | [video] |
| Regular Walk w/ Obstacle | 2 | None | No | Yes | [video] |
| Limp | 3 | 1 (left leg) | No | No | [video] |
| Limp w/ Gait Freeze | 4 | 1 (left leg) | Yes | No | [video] |
| Limp w/ Obstacle | 5 | 1 (left leg) | No | Yes | [video] |
| Shuffle | 6 | 2 (both legs) | No | No | [video] |
| Shuffle w/ Gait Freeze | 7 | 2 (both legs) | Yes | No | [video] |
| Shuffle w/ Obstacle | 8 | 2 (both legs) | No | Yes | [video] |
File Structure and Usage Instructions
The make-up of the various downloadable files available here including content and file structure:
Samples
Please see the Sample Data section above for access to individual videos from the dataset. There are a total of 9 sample videos, each denoting one of each factor combination as described in the table below. Each video features a different participant and the background visibile in every video is the same background used for every participant in the dataset.
Variable Combinations and Counts per Participant.
| None | Freeze | Obstacle | |
|---|---|---|---|
| Regular | 10 | 5 | 5 |
| Limping | 10 | 5 | 5 |
| Shuffling | 10 | 5 | 5 |
Full Data Download
These are direct download links for a complete dataset download (courtesy University of Edinburgh DataShare). Please refer to section Download Complete Dataset below. The full downloadable data contains four zip files. The nested list show the naming conventions of the files in each of the zip file.
- All RGB Videos: All RBG videos in raw form without pre-processing.
- All Depth Videos: All Depth videos in raw form without pre-processing.
- All Joints: All joints extracted from original raw data in .csv form. See the Readme below for formatting information.
- Example Videos: the 9 video examples shown above.
A Readme.md file explaining the example file structure of individual CSV files is available in the full dataset.
Download the Complete Dataset
Download Combined File Size: 1.7GbRelated Publications
All publications using data from the "WeightGait" dataset should cite the following paper:
- C. Lochhead, R. B. Fisher; A Lightweight Approach to Gait Abnormality Detection for At-Home Health Monitoring, Computers in Biology and Medicine, vol 190, 2025.
Acknowledgements
Firstly, we would like to thank the dataset developers Christopher Lochhead and Robert B. Fisher. We would also like to thank the Advanced Care Research Centre (ACRC) for their funding of both the PhD studentship and data collection. We'd also like to thank all of the participants who took part in the study and dedicated their time to help us develop this valuable resource.
Terms and Conditions of Use
- This dataset can be used for academic purposes only.
- For the privacy of subjects, their names are replaced with a subject number, by which each of these are referred to in the dataset. Additionally, images of the subjects in the dataset are only allowed for demonstation in academic publications.
- Any use of the dataset in a publication must cite: C. Lochhead, R. B. Fisher; A Lightweight Approach to Gait Abnormality Detection for At-Home Health Monitoring, Computers in Biology and Medicine, vol 190, 2025.
There have been
accesses since April 2025.