Pedometer Evaluation Project
Problem
Pedometers have become ubiquitous devices for helping motivate people
to be more active by counting their daily steps.
However, pedometer accuracy varies widely depending on the body
position at which it is worn, the signal processing algorithm used
to count steps, and individual differences in human gait.
This project recorded a large data set of raw accelerometer data and
marked the time occurrences of all steps so that pedometer algorithms
could be evaluated objectively against a gold standard.
Data
The below image shows raw (x,y,z) accelerometer data for 3 different
devices worn on the wrist, hip and ankle (top to bottom), along with
a screenshot of the synchronized video recording.
Click the image to watch a video snippet.
 |
 |
| Method used to record data and ground truth |
Example data (click to watch video snippet) |
A group of 30 subjects were recorded walking a cumulative total of over
60,000 steps, including portions in a regular gait (walking a path),
semi-regular gait (walking around a building), and irregular gait
(walking around a room).
UPDATE Nov 2022: A pre-compiled program (includes source code) has been added
that allows you to view the synchronized video and data (see link below).
- The data set and ground truth can be downloaded
here (31 MB).
In a data file, the 9 columns are wrist xyz, hip xyz and ankle xyz.
In a ground truth file, the number indicates the line index in the data file
at which a step occurred. Data has been synrhonized between all 3 sensors and
resampled to exactly 15 Hz.
-
Video files can be downloaded
here (11 GB).
-
The original raw sensor recordings (unsynchronized, slightly asynchronous)
that include accelerometer, gyroscope and magnetometer data can be downloaded
here (277 MB).
-
A Windows program that displays the data, video and ground truth can
be downloaded
here (43 MB).
The program installer includes the source code and auxilary files for
compiling it using MS Visual Studio.
-
Demographics of participants including gender, age, height, weight, handedness.
Papers about this project:
-
R. Mattfeld, E. Jesch and A. Hoover,
"A new dataset for evaluating pedometer performance",
in the proc. of 2017 IEEE International Conference on
Bioinformatics and Biomedicine, Kansas City, MO, pp. 865-869, 2017.
-
R. Mattfeld,
"Evaluation of Pedometer Performance Across Multiple Gait Types
Using Video for Ground Truth",
PhD dissertation, Electrical and Computer Engineering Department,
Clemson University, 2018.
-
R. Mattfeld, E. Jesch and A. Hoover,
"Evaluating Pedometer Algorithms on Semi-Regular and Unstructured Gaits",
in Sensors, vol. 21 no. 13, June 2021, doi: 10.3390/s21134260.
Pedometer Evaluation Page / Clemson / ahoover@clemson.edu
