Datasets

The datasets published by the lab are organised according to their project group affiliation.

Applied Machine Learning

Nuremberg Polar Bear Dataset

Description

The images have been taken at the polar bear enclosure at Nuremberg Zoo, which is home to two mature animals (Vera, female adult and Nanuq, male adult). The three on-site cameras acquire videos with a frame rate of 12.5 fps and a resolution of 3840×2160 pixels. For the aim of this project, a period of five days of data (27 April – 1 May 2020) has been selected. During this period, the polar bears shared both enclosures and thus might both be present in a single image. A total of 4450 frames were randomly selected and stored for further labeling. Three biologists annotated all images to provide labels of high quality by assigning labeled bounding boxes to the animals visible in the picture. We provide the resulting mean annotations. As the 4450 images were randomly selected from the video data, only 2099 instances show one or more animals. 167 images show both animals, 1932 only one. 2266 bounding boxes are provided, 1082 for the male and 1184 for the female bear.
For further information, please refer to the associated paper.

Publication

Please cite this publication when using the dataset:

Download Link

To download the dataset, please refer to the official zenodo website.

Dataset Coordinator

Matthias Zürl (M. Sc.)

 

Biomechanical Motion Analysis and Creation

Dataset for Metabolic Cost Calculations of Gait using Musculoskeletal Energy Models, a Comparison Study

Description

This data set contains raw and processed data of gait analysis experiments of level and inclined walking at two speeds for 12 participants. The slopes were uphill and downhill with 8% incline. The raw data contains the output of the force plates and marker data, as well as raw measurements from an K4B2 system. Mat files are processed data: measured metabolic rate, and measured and calculated metabolic cost, as well as kinetic and kinematic data of an averaged gait cycle: joint angles, velocities and moments, ground reaction forces, muscle activation, contractile element length and stimulation, and the duration of the gait cycle.
For further information, please refer to the official zenodo website.

Publication

Please cite this publication when using the dataset:

Download Link

To download the dataset, please refer to the official zenodo website.

Dataset Coordinator

Prof. Dr. Anne Koelewijn

RunMaD: Efficient simulation of 3D musculoskeletal model with implicit dynamics

Description

In this dataset, a full-body three dimensional musculoskeletal model is extended to be specialized for running with directional changes. Model dynamics were implemented implicitly and trajectory optimization problems were solved with direct collocation to enable efficient computation. Standing, straight running, and curved running were simulated starting from a random initial guess to confirm the capabilities of our model and approach: efficacy, tracking and predictive power. Altogether the simulations required 1 h 17 min and corresponded well to the reference data.
For further information, please refer to the associated publication.

Publication

Please cite this publication when using the dataset:

Download Link

To download the dataset, please refer to the official SimTK website.

Dataset Coordinator

Marlies Nitschke (M. Sc.)

Optical motion capturing of change of direction motions reconstructed with inverse kinematics and dynamics and optimal control simulation

Description

This study investigated the feasibility and accuracy of reconstructing, especially change of direction motions with a 3D full-body musculoskeletal model by tracking marker and ground reaction force (GRF) data in optimal control simulations. We recorded in total 30 trials with optical motion capture. Using this data, we compared inverse methods (inverse kinematics and dynamics) to coordinate tracking simulations and marker tracking simulations.
For further information, please refer to the associated publication.

Publication

Please cite this publication when using the dataset:

Download Link

To download the dataset, please refer to the official zenodo website.

Dataset Coordinator

Marlies Nitschke (M. Sc.)

 

Digital Health – Biosignals

Scaled and Translated Image Recognition (STIR)

Description

While convolutions are known to be invariant to (discrete) translations, scaling continues to be a challenge and most image recognition networks are not invariant to them. To explore these effects, we have created the Scaled and Translated Image Recognition (STIR) dataset. This dataset contains objects of size s[17,64], each randomly placed in a 64×64 pixel image.
For further information, please refer to the official zenodo website.

Publication

Please cite this publication when using the dataset:
[1] Altstidl, T., Nguyen, A., Schwinn, L., Köferl, F., Mutschler, C., Eskofier, B., & Zanca, D. (2022). Just a Matter of Scale? Reevaluating Scale Equivariance in Convolutional Neural Networks. arXiv preprint arXiv:2211.10288.

Download Link

To download the dataset, please refer to the official zenodo website.

Dataset Coordinator

Thomas Altstidl (M. Sc.)

DaLiAc - Daily Life Activities

Description

The DaLiAc (Daily Life Activities) database consists of data from 19 subjects (8 female and 11 male, age 26 ± 8 years, height 177 ± 11 cm, weight 75.2 ± 14.2 kg, mean ± standard deviation (SD)) that performed 13 daily life activities. These activities were chosen according to their different Metabolic Equivalent of Task (MET) values. Four SHIMMER (Shimmer Research, Dublin, Ireland) sensors were used for data acquisition. Each sensor node was equipped with a triaxial accelerometer and a triaxial gyroscope. Data were sampled with 200 Hz and were stored on SD card. The sensor nodes were placed on the left ankle, the right hip, the chest, and the right ankle.
For further information, please refer to the associated publication.

Publication

Please cite this publication when using the dataset:

Download Link

The download link is currently not available, please contact the dataset coordinator. For further information, please visit the official MaD Lab website.

Dataset Coordinator

Dr.-Ing. Heike Leutheuser

BaSA - Basic Step Activities

Description

Two SHIMMER (Shimmer Research, Dublin, Ireland) sensors were used for data acquisition. Each sensor node was equipped with a triaxial accelerometer and a triaxial gyroscope. Data were sampled with 200 Hz and were stored on SD card. The Basic Step Activities database consists of data from 15 subjects (8 female and 7 male, age 23 ± 2 years, height 178 ± 12.5 cm, weight 75.0 ± 16.5 kg, mean ± standard deviation (SD)) that performed 7 daily life activities. These activities were chosen according to their different Metabolic Equivalent of Task (MET) values.
For further information, please refer to the associated publication.

Publication

Please cite this publication when using the dataset:

Download Link

The download link is currently not available, please contact the dataset coordinator. For further information, please visit the official MaD Lab website.

Dataset Coordinator

Dr.-Ing. Heike Leutheuser

EnEx - Energy Expenditure

Description

The EnEx (Energy Expenditure) database consists of data from ten subjects (3 female and 7 male, age 49 ± 12 years, height 178 ± 10 cm, weight 80.7 ± 14.6 kg, mean ± standard deviation (SD)). In one trial, each subject had to run on a traditional treadmill at three different speed levels ([3.2, 4.8, 6.4] km/h). In a second trial, an oscillating treadmill was used imposing different levels of physical activity. Each speed level lasted six minutes. Two SHIMMER sensor nodes were placed on right hip and right ankle. Each sensor node consisted of three accelerometer axes and three gyroscope axes. The sampling rate of the sensor nodes were set to 204.8 Hz. For the expended energy, the oxygen consumption was measured by a spirometry system. The sampling rate was 0.2 Hz.
For further information, please refer to the associated publication.

Publication

Please cite this publication when using the dataset:

Download Link

The download link is currently not available, please contact the dataset coordinator. For further information, please visit the official MaD Lab website.

Dataset Coordinator

Dipl.-Ing. Dominik Schuldhaus

 

Digital Health – PsychoSense

Cold Face Test Dataset

Description

In this study, the use of the Cold Face Test (CFT) protocol as an intervention to reduce acute stress responses was investigated. Twenty-eight healthy participants were exposed to acute psychosocial stress via the Montreal Imaging Stress Task (MIST) in a randomized between-subjects design while heart rate (HR), heart rate variability (HRV), and salivary cortisol were assessed.
For further information, please refer to the associated publication

Publication

Please cite this publication when using the dataset:

Download Link

To download the dataset, please refer to the official OSF website.

Dataset Coordinator

Robert Richer (M. Sc.)

CARWatch Dataset

Description

This dataset contains cortisol awakening response (CAR) data, objective saliva sampling time logging from the CARWatch smartphone application, and IMU-based movement during the night. It was used as proof-of-concept validation of the CARWatch application for objective sampling time assessment and to assess the influence of the inner clock on the CAR and pre-awakening movement. For further information, please refer to the associated publications (see below).

Publication

Please consider citing these publications when using the dataset:

Download Link

To download the dataset, please refer to the official OSF website.

Dataset Coordinator

Robert Richer (M. Sc.)

Digital Health – Gait Analytics

Sensor Position Comparison

Description

A dataset containing IMU recordings with full motion capture reference from 14 participants (approx. 10000 strides). Each participant was equipped with 15 synchronised IMUs (6 at different positions at each shoe, 1 at each ankle, and 1 and the lower back). The main goal of the dataset is to compare the recorded signals of the 6 sensors attached to each foot.
For further information, please refer to the ‘README.md’ file linked on the official zenodo website.

Publication

Please cite this publication when using the dataset:

Download Link

To download the dataset, please refer to official zenodo website.

Dataset Coordinator

Arne Küderle (M. Sc.)

FallRiskPD Dataset (summary gait parameters)

Description

This dataset provides spatio-temporal gait parameters recorded from 35 Parkinson’s disease patients during real-world gait and unsupervised 4×10 Meter Walking tests.

The data was collected as part of the FallRiskPD study (DRKS-ID: DRKS00015085) between March 2019 and June 2021 by the University Hospital Erlangen, the Hospital Rummelsberg, and the Ernst von Bergmann Hospital Potsdam. Raw data was recorded with the Mobile GaitLab (Portabiles HealthCare Technologies GmbH, Erlangen, Germany), consisting of two foot-worn inertial measurement units placed to the instep of the shoes. The recordings were acquired over the course of approximately two weeks, while the participants were following their activities of daily living and were additionally asked to perform 4×10 Meter Walking tests three times per day with preferred walking speed. After the gait recordings, fall events were captured over three months in a paper-based diary.

The sensor recordings were processed to stride-level spatio temporal gait parameters using a pipeline of gait sequence detection (Ullrich et al., 2020, JBHI), stride segmentation (Barth et al., 2015, Sensors), and event detection with trajectory reconstruction (Rampp et al., 2015, TBME). The following gait parameters are available: stride time [s], stance time [s], swing time [s], stride length [m], gait speed [m/s], IC foot angle [deg], FC foot angle [deg], maximum foot lift [m].

Three files are provided with the data set:

  1. strides_realworld_publication.csv: Contains the parameterized strides of the real-world activities including the patient ID, the date and time information, and the IDs of the corresponding walking bouts in which the respective strides can be grouped together.
  2. strides_4x10m_pref_at_home_publication.csv: Contains the parameterized strides of the unsupervised 4×10 Meter Walking tets including the patient ID, the date and time information, and the IDs of the corresponding gait tests that are counted with a suffix in the ID within one day.
  3. patient_data_publication.csv: Contains the basic patient information including demographics, UPDRS-III score, Hoehn and Yahr stage, and the occurance of falls in the follow up phase.

For further information, please refer to the associated publication and the official OSF website.

Publication

Please cite this publication when using the dataset:

Download Link

To download the dataset, please refer to the official OSF website.

Dataset Coordinator

Martin Ullrich (M. Sc.), Arne Küderle (M. Sc.)

Stair Ambulation Dataset

Description

As stairs are an essential part of our everyday lives and are frequently encountered in urban environments, stair ambulation sequences should be included in mobility analysis. Therefore, this dataset was recorded in January 2021 to enable the development and evaluation of algorithms for human gait analysis using wearable inertial sensors for real-world applications, including level-walking gait as well as stair ascending and stair descending.
For further information, please refer to the associated publication.

Publication

Please cite this publication when using the dataset:

Download Link

To download the dataset, please refer to the official OSF website.

Dataset Coordinator

Nils Roth, Arne Küderle (M. Sc.)

Sensor-based Gait Analysis Validation Data

Description

The database consists of data from 15 subjects (eleven healthy subjects and four PD patients) that performed the 4×10 m straight walking tests. Data was acquired using two Shimmer3 (Shimmer Research, Dublin, Ireland) sensors laterally attached to the shoes. Data from 3D accelerometer and gyroscope were transmitted to a mobile device. The reference system was a camera-based markerless motion capture (Simi, Unterschleißheim, Germany). Synchronization was assured at the beginning and end of the measurements and the data was aligned accordingly.
For further information, please refer to the associated publication.

Publication

Please cite this publication when using the dataset:

Download Link

https://osf.io/cfb7e/

Dataset Coordinator

Dr.-Ing. Felix Kluge

GaitPhase Database

Description

The GaitPhase database consists of gait data from 21 subjects (10 male, 11 female, age: 23.8 yrs ± 3.3 yrs, height: 172.8 cm ± 9.4 cm, weight: 66.6 kg ± 10.9 kg; all values are mean ± standard deviation). In total, 25306 steps were acquired. The performed excercise was walking on a split-belt treadmill at 12 different speeds in the interval [0.6, 1.7] m/s with 0.1 m/s increments for one minute at each speed.
Three-dimensional kinematic data using a Qualisys (Qualisys AB, Gothenburg, Sweden) motion capture system with 8 Oqus cameras sampling at 200 Hz and an instrumented Bertec (Bertec Corporation, Columbus, OH, USA) split-belt treadmill with integrated force plates sampling at 1000 Hz were used for data acquisition. Both systems were synchronized camera frame-wise.
For further information, please refer to the associated publication.

Publication

Please cite this publication when using the dataset:

Download Link

The download link is currently not available, please contact the dataset coordinator. For further information, please visit the official MaD Lab website.

Dataset Coordinator

Dr.-Ing. Felix Kluge

Benchmark cyclic activity recognition database using wearables

Description

The dataset consists of 12 different task-driven activities, 10 of which are cyclic. It contains over 150,000 labeled cycles, each with 2 phases, from 80 subjects. The activities include not only straight and steady-state motions, but also transitions, different ranges of bouts, and changing directions. Each participant wore 5 synchronized inertial measurement units (IMUs) on the wrists, shoes, and in a pocket, as well as pressure insoles and video.
For further information, please refer to the associated publication.

Publication

Please cite these publications when using the dataset:

Download Link

The download link is currently not available, please contact the dataset coordinator. For further information, please visit the official MaD Lab website.

Dataset Coordinator

Christine Martindale (M. Sc.)

Wearable multi-sensor gait-based daily activity data

Description

The database consists of data from 20 healthy subjects. Their characteristics are as follows: 5 females and 15 males, with an average age of 28 years, an average height of 175 cm and weight of 74 kg. The shoe sizes were limited to the range of 38 to 44 due to the available insole sizes. Data was acquired using five IMU sensors laterally attached to the left and right shoes as well as wrists and within the right pocket. Data from 3D accelerometer and gyroscope were recorded and well as pressure data from Moticon insoles. The data was synchronised and labelled using the smart annotation tool.
For further information, please refer to the associated publication.

Publication

Please cite this publication when using the dataset:

Download Link

The download link is currently not available, please contact the dataset coordinator. For further information, please visit the official MaD Lab website.

Dataset Coordinator

Christine Martindale (M. Sc.)

Smart Annotation Cyclic Activities Dataset

Description

The database consists of data from 18 healthy subjects who performed walking and running. The participants performed walking and running and standing in an outdoor environment on varying surfaces and performed a small circuit around the building. Data was acquired using two IMU sensors laterally attached to the left ankle. Data from 3D accelerometer and gyroscope were recorded. The data was annotated by hand using a camera reference.
For further information, please refer to the associated publication.

Publication

Please cite this publication when using the dataset:

The download link is currently not available, please contact the dataset coordinator. For further information, please visit the official MaD Lab website.

Dataset Coordinator

Christine Martindale (M. Sc.)

eGaIT - embedded Gait analysis using Intelligent Technologies

Description

The eGaIT database for the stride segmentation validation consists of data from 70 subjects. 25 elderly controls for template generation, 15 elderly controls and 15 patients with Parkinson’s disease (PD) for algorithm evaluation were recorded at the Movement Disorder Unit of the Department of Molecular Neurology, University Hospital Erlangen, Germany. Sensor data from 15 geriatric patients were acquired at the Geriatrics Centre, Waldkrankenhaus St. Marien, Erlangen, Germany. The eGaIT database for the gait parameter validation consists of data from 101 subjects (55 female and 46 male, age 82.1 ± 6.5 years, height 164.0 ± 10.0 cm, mean ± standard deviation (SD)) that performed straight walking tests. Two SHIMMER (Shimmer Research, Dublin, Ireland) sensors were used for data acquisition. Each sensor node was equipped with a triaxial accelerometer and a triaxial gyroscope and was attached to the lateral side of a sports shoe. Data were sampled with 102.4 Hz and were streamed via Bluetooth to a standard windows computer.
For further information, please refer to the associated publication.

Publication

Please cite these publications when using the dataset:

Rampp, A., Barth, J., Schülein, S., Gaßmann, K. G., Klucken, J., & Eskofier, B. M. (2014). Inertial sensor-based stride parameter calculation from gait sequences in geriatric patients. IEEE transactions on biomedical engineering, 62(4), 1089-1097.

Download Link

The download link is currently not available, please contact the dataset coordinator. For further information, please visit the official MaD Lab website.

Dataset Coordinator

Dr. Jens Barth, Alexander Rampp

 

Empatho-Kinaesthetic Sensor Technology (EmpkinS)

 

Sports Analytics