Pedometer Evaluation Dataset

This dataset designed to improve the ability for researchers to evaluate pedometer algorithms. The dataset allows pedometer algorithms to be evaluated based on a long term running count accuracy (RCA) and on the algorithm’s individual step detection accuracy (SDA). This level of detailed analysis is made possible with video recordings of each activity performed in the experiment, allowing for objective ground truth determination which can be objectively reviewed. In addition to data collection across multiple sensors located in three positions across the body, the dataset collects data from three different activities, each designed to elicit a unique gait type common in free living. In total, 30 participants and over 60,000 steps were evaluated.

The dataset, downloadable below, contains a ReadMe file describing the contents and data included. The dataset contains a demographics file identifying height, weight, gender, and handedness information for each participant. A file containing Fitbit step counts before and after each activity for each participant is also provided. Three folders are located within each participant folder, each containing data from one of the three activities performed. Within each activityh folder, a variety of data collected through Shimmer3 sensors located on the wrist, hip, and ankle are provided. The start times for the data within each activity folder were synchronized across each sensor position. The data was captured at a rate of 15 Hz. A sync file indicates the video file name and the number of indices the data must be shifted in order to synchronize with the video recording. The steps.txt folder in each file indicates the ground truth step times in indices (where each index is 1/15 of one second), as well as foot used to take the step (r for right foot and l for left foot). Shifts in the dataset are indicated by rshift for shifts taken with the right foot and lshift for shifts taken with the left foot.

The dataset is described in detail in two publications:

R. Mattfeld, “Evaluation of Pedometer Performance Across Multiple Gait Types Using Video for Ground Truth,” PhD dissertation, Holcombe Department of Electrical and Computer Engineering, Clemson University, SC., 2018.

R. Mattfeld, E. Jesch and A. Hoover, "A new dataset for evaluating pedometer performance," 2017 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Kansas City, MO, pp. 865-869, 2017.

The data is available for download here: 

https://cecas.clemson.edu/~ahoover/pedometer/ 

Including:

1) synchronized data

2) video files

3) the original, raw, unsynchronized data

4) a Windows program to display the data, video, and ground truth.

The original, raw, unsynchronized data can also be downloaded using the links below: 

Acceleromter data and step labels

Video and synchronization information