Simulated gait dataset for abnormality detection

The dataset is created by collecting volunteers walking data using inertial motion units (IMU) placed on the forefoot. Data collection device was Shimmer S3 IMU.
Volunteers simulated eight different common gait deviations focusing on lower extremities.
Simulation was recreating actual patients' gait patterns, under the instructions of professional physiotherapist.
Each gait recording contains normal ("ok") and abnormal ("ab") step patterns, as well as turn ("turn") steps.
Start of the recording is shown by label "start" and "stop" of the recording is shown by label stop.
In total, there are 22 different persons, who simulated different combinations of abnormal step patterns. 
Data was annotated by the definitive shapes of step patterns using gyroscope magnitude and gait recording videos for reference.

In the main folder of "depersonalyzed_data" are folders containing data for each person separately.
Then for each person folder, there are folders named after gait types, which were simulated.
In each gait type folder, there are one to several gait recordings in CSV files, and corresponding labels in the txt files.
The delimiter in the CSV file is tab. In the TXT file, each label is on the new line.

CSV file has the following structure. Columns: time in ms, accelerometer axis x,y,z in m/(s^2) and gyroscope axis x,y,z in deg/s. Rows: first row is containing standard Shimmer values, which are unique for each column.
Second row contains units of measurement, starting from third row is data itself, where decimal is separated by full stop.
Sampling rate is set to 256Hz. Accelerometer range is 8G and gyroscope range is 1000 deg/s.

TXT file has the following structure. Columns: start of the label, end of label and value of the label.

Data has been used in the publications listed below, first for classification and then for real-time in-step anomaly detection.
The studies approved by Estonian Research Ethics Committee of the National Institute for Health Development, permission No.818. The participants provided their written informed consent to participate in this study.

To access the data please use link below:
https://drive.google.com/drive/folders/10FDCG4xpRwhuYfh9eatbPI0D0JsHxOe_?usp=sharing

本数据集通过采集志愿者行走数据构建，采集时将惯性运动单元（Inertial Motion Unit, IMU）佩戴于前足。数据采集设备为Shimmer S3 IMU。
志愿者模拟了8种常见的下肢步态异常模式。所有模拟动作均在专业物理治疗师的指导下进行，复刻真实患者的步态模式。
每段步态录制数据均包含正常（标注为"ok"）、异常（标注为"ab"）步型，以及转向（标注为"turn"）步态。数据录制的起始和结束分别由标签"start"和"stop"标识。
本次数据集共纳入22名志愿者，他们分别模拟了多种异常步型的组合。
数据标注以陀螺仪幅值特征与步态录制视频作为参考依据，通过步型的典型特征完成标注。

在名为`depersonalyzed_data`的主文件夹中，包含按每位志愿者单独划分的数据子文件夹。每个志愿者文件夹内，均包含以其所模拟的步态类型命名的子文件夹。每个步态类型子文件夹中，包含1至多段以CSV格式存储的步态录制数据，以及对应的TXT格式标签文件。CSV文件的分隔符为制表符；TXT文件中，每条标签各占一行。

CSV文件的结构如下：列字段依次为时间（单位：毫秒）、加速度计x/y/z轴数据（单位：米每二次方秒，m/s²）、陀螺仪x/y/z轴数据（单位：度每秒，deg/s）。文件首行包含各列对应的标准Shimmer设备参数，每列参数唯一；第二行为各字段的计量单位；从第三行起为原始采集数据，小数点采用英文句号分隔。采样率设置为256Hz；加速度计量程为8G，陀螺仪量程为1000 deg/s。

TXT标签文件的结构如下：每一行包含标签起始时间、标签结束时间以及标签值。

本数据集已用于以下发表研究，先后被应用于步态分类与实时步内异常检测任务。本研究已通过国家卫生发展研究院爱沙尼亚研究伦理委员会审批，审批编号为818。所有参与者均已签署书面知情同意书。

如需获取该数据集，请使用以下链接：https://drive.google.com/drive/folders/10FDCG4xpRwhuYfh9eatbPI0D0JsHxOe_?usp=sharing

搜集汇总

背景与挑战

背景概述

该数据集是通过IMU设备收集的22名志愿者模拟不同异常步态的数据，包含正常、异常和转身步态模式，数据以CSV和TXT格式存储，采样率为256Hz。数据集主要用于步态异常检测研究，并已用于多项相关出版物中。