Dem@Care Datasets

Dem@Care is providing the following datasets, which are collected during lab and home experiments. The data collection took place in the Greek Alzheimer’s Association for Dementia and Related Disorders in Thessaloniki, Greece and in participants' homes. The datasets include video and audio recordings as well as data from physiological sensors. Moreover, they include data from sleep, motion and plug sensors.
These datasets are available to the research community under specific terms of use.

Dem@Care Datasets


Dataset Name




Video files from static, color-depth cameras (Asus RGB-D sensor, Kinect)



Video files from wearable camera (GoPro)



Audio files from microphone



Data from physiological sensors

5 Ds5 Audio files from microphone from Dem@Care short @Lab protocol
6 Ds6 Video files from static, color-depth cameras (Asus RGB-D sensor, Kinect) from Dem@Care short @Lab protocol
7 Ds7 Audio files from microphone from Dem@Care long @Lab protocol
8 Ds8 Video files from static, color-depth cameras (Asus RGB-D sensor, Kinect) from Dem@Care long @Lab protocol
9 Ds9 A) data from motion sensor on objects B) data from plug sensor C) data from physiological sensor (DTI-2) from Dem@Care long @Lab protocol 
10 Ds10 A) data from motion sensor on objects B) data from plug sensor C) data from physiological sensor (DTI-2) from Dem@Care short @Lab protocol
11 Ds11 A) data from motion sensor on objects B) data from plug sensor C) data from physiological sensor (UP24) D) data from sleep sensor (AURA)  from Dem@Care 1st home pilot 
12 Ds12 A) data from motion sensor on objects B) data from plug sensor C) data from physiological sensor (UP24) D) data from sleep sensor (AURA)  from Dem@Care 2nd home pilot

In order to provide you part of the Dem@Care datasets you have to follow the following steps: 

  1. Read the datasets Terms of Use below which oblige you, among other things, to:
       not share this data with anyone else

       not attempt to identify any of the individuals

  1. Choose which datasets you are going to need in your research.
  2. Prepare a title and a short description (abstract) of your project.
  3. Send us an email (based on the following template) using your institution’ s email. In order to grant you access to the data we need it to contain the template statements.

Before you apply:
  1. Be as specific as possible about why you want each data table. 
  2. In the case of undergraduate and postgraduate students, you should ask your supervisor to complete the form below, and mention you as a supervisee. We cannot collaborate with researchers who do not have a full-time academic contract.
  3. If you intend to ask for an alternative collaboration agreement from the default one, explain why you think this is appropriate.
  4. The datasets Terms of Use are serious. Please read them all. If you do not follow them we will require you to delete all Dem@Care data, and may ask journals to retract any publications from the period when you did not follow them. In extreme cases we will contact university departments where collaborators work.

Email template:

I would like to become a registered collaborator of the Dem@Care project to pursue the following research: 


I require access to the following datasets:


I am planning to use the following variables: 


I agree with the Dem@Care datasets Terms of Use, and will take responsibility for the use of the data by any students in my research group.

I also agree that Dem@Care can include my name, my position, my institution, and the title of my research topic on its public Web page and other documents. If I cease work on Dem@Care data, then this collaboration will be listed with an ending date. 

I agree that there will be a reference to: Karakostas A., Briassouli A., Avgerinakis K., Kompatsiaris I., Tsolaki M. "The Dem@Care Experiments and Datasets: a Technical Report”, arXiv:1701.01142 [cs.CV],

Your Name / Position / Institution 

Datasets Terms of Use


You cannot send the datasets to any other party (even if they have access to it themselves), nor disclose to anyone else the information contained within it as well as its structure. It is allowed to share data with students in your research group who you agree to supervise and take full responsibility that their use of the data also meets these Terms of Use.


You should not link individual data (records, image or video files etc) with any other information about an individual that you may have. This implies that you cannot attempt to contact any individuals either.

Moreover, if you are going to use images or video captures from Dem@Care datasets in any of your papers or presentations, in any case you should not show patients’ faces.  

Non-Commerical License

We grant you a non-commercial license to use the data. You can only use it for academic research that does not earn revenue, and your research also cannot be in collaboration with any commercial entities.

Scope of Research

If the scope of your research changes, then you should contact us again for us to agree the change. If you stop researching using Dem@Care data you should make us aware of this fact.

Publication Reference

On any publications that arise from Dem@Care data you should reference to: Karakostas A., Briassouli A., Avgerinakis K., Kompatsiaris I., Tsolaki M. "The Dem@Care Experiments and Datasets: a Technical Report”, arXiv:1701.01142 [cs.CV],

Ethical Clearance

Although, all of Dem@Care tests require that users read and agree to information regarding using their data for research, making the data available to other researchers, and their ability to withdraw from the research at any time, it is your responsibility to ascertain whether it is ethical to use the data that Dem@Care has collected for academic research.

Acknowledgement of Dem@Care

All uses of Dem@Care data, including but not limited to research papers, at conferences, on websites, and in press releases, should include prominent acknowledgement that Dem@Care is the data source. However, it should not imply that Dem@Care endorses the research. It should be clear that Dem@Care is an external data supplier.

Access to data

Dem@Care cannot guarantee continued access to the data, nor that our service will not be interrupted from time to time. The license to use and store our data is recoverable, which means that Dem@Care may ask you to cease use of it and to delete it from any storage you have at our sole discretion.


Dem@Care disclaims any warranties, for example but not limited to, the data's suitability for research or publication





Research objectives

Datasets downloaded


Kingston University

Prof. Vasileios Argyriou

This project is focused on designing and implementing mechanisms to monitor the patients’ behaviour in a non-intrusive way during their everyday life considering all the related ethical issues will be introduced.

The depth information and the position of the joints will be used. Features are going to be defined and this information will be used to train a machine learning algorithm that will detect different human behaviours. 



University of Malaya

Prof Loo Chu Kiong

Our project focuses on evaluating such features to assess its performance in ADL recognition for both first and third person perspective. Another novel approach is to study the fusion of both egocentric (first person) and third person viewpoint for ADL recognition. The Dem@Care dataset will be used to evaluate the performance of the proposed method. The dataset also allow us to investigate algorithms that discriminate between activity performed by a healthy person and person with mild dementia.

DS1, DS2


MIT, Lincoln Laboratory

Bea Yu

Associate Technical Staff

We have recently used speech and video features to automatically predict depression [1,2], winning both AVEC 2013 and 2014 Depression Sub-Challenges.  Given the potential health care implications of rising numbers of people suffering from dementia due to a large aging population in the US and around the world, we have expanded our focus to automated dementia prediction and monitoring using speech and features from other modalities [3].  The Dem@Care Datasets appear to be a very useful resource for this purpose and we were very excited to see that they are available to academic researchers.



Department of Computer Science

FAST National University, Karachi

Furqan M Khan

Assistant Professor

Supervised recognition of daily living activities



Università degli Studi di Milano

Prof. Claudio Bettini

A non-intrusive sensor-based infrastructure acquires low-level data about the interaction of the individual with the home environment including objects, appliances and furniture. Our goal is to detect abnormal behaviors at a fine-grained level, thus providing an important tool to support the medical diagnosis.

DS9, DS10, DS11, DS12


Faculty of Electronics, Telecommunications and Information Technology University Politehnica of Bucharest

Assoc. Prof. Dr. Eng. Bogdan IONESCU

The research proposes to implement a solution in order to monitor and track down the needs of old and sick people, based on advanced video processing. 

The project will test different scenarios of and using skin detection and other techniques will try to find the best method in distinguish a healthy person from a sick one . All these methods and techniques will be implemented in MatLab environment.

DS1, DS3, DS6


Biomedical group at Mondragon University (Spain)

Dr. Asier Aztiria

The general objective is to develop a system that learns how patients’ behaviours and biomedical signals are affected when a patient suffers a specific disorder. Such knowledge will be used to detect symptoms in a transparent way and to help in the early diagnosis of such disorders.

DS1, DS3, DS4, DS9,
DS10, DS11, DS12


Faculty of Engineering, Computing and Science

Swinburne University of Technology Sarawak Campus

Dr. Lau Bee Theng

Associate Professor

The aim of the project is to develop an affordable robot for children, the elderly and disabled patients that helps to autonomously monitor for possible injuries while providing an avatar for Telepresence by the carer, in addition to being a highly extensible assistive robot development platform for every home

DS1, DS2, DS6, DS8


University of Bristol

Yangdi Xu

A person’s routine incorporates the frequent and regular behaviour patterns over a time scale, e.g. daily routine. In this work we present a method for unsupervised discovery of a single person’s daily routine within an indoor environment using a static depth sensor. Routine is modelled using top down and bottom up hierarchies, formed from location and silhouette spatio-temporal information. We employ and evaluate stay point estimation and time envelopes for better routine modelling. The method is tested for three individuals modelling their natural activity in an office kitchen. Results demonstrate the ability to automatically discover unlabelled routine patterns related to daily activities as well as discard infrequent events.



Department of Computer Science & Engineering,
SriVenkateswara College of Engineering Sriperumbudur

Associate Professor

The project focuses on development of machine learning based algorithms for the automated recognition of daily activities of older people. The datasets could help substantially with the development and evaluation of algorithms for understanding the behavior of the people with and without dementia.

DS1, DS2, DS6, DS8

11 German Research Center for Artificial Intelligence Nicklas Linz Project: ELEMENT - Early Detection of Cognitive Disorders such as Dementia on the Basis of Speech Analysis
Abstract: The project is focused on developing a light screening application to access a persons cognitive health through speech analysis.
The data form Dem@Care will be used to verify the accuracy of machine learning models trained on other resources.
DS5, DS7
12 Institut Supérieur d'Informatique et Multimédia, (
Université de Sfax
Ing. Yassine Ben Ayed
Associate Professo
Acoustic variables from the audio files,and linguistic variables from the associated transcripts, and use these variables to train a machine learning classifier to distinguish between participants with AD and healthy controls. DS5, DS7
13 Monta Vista High School in Cupertino, California, USA Renee Fallon The Voice of Alzheimer’s: Wearable Technology Coupled with Machine Learning to Track Alzheimer’s Disease Progression

Alzheimer’s disease is the most prevalent form of dementia throughout the world, and despite significant advancements in treatment, tracking disease progression remains a challenge. To address this, his project seeks to use wearable technology in conjunction with machine learning, voice recognition/detection, and signal processing algorithms to create a comprehensive, quantitative method of analyzing the unique vocal aspects that vary from patient to patient in order to effectively track AD progression.
DS3, DS5, DS7
14 University Utara Malaysia Prof Abdull Sukor Shaari This project focuses on investigating elderly people's behaviour based on their daily routine and identifying an abnormal situation. The aim is to develop an efficient reasoning system that can monitor elderly individual's daily lives and detect abnormalities as well as identify the most probable reason and solution. The system is supported by a semantic knowledge base that can represent knowledge of the world in order to support the reasoning system. DS4, DS5, DS10, DS11, DS12
15 De Montfort University Ismini Psychoula The project focuses on designing and developing mechanisms that help residents of ambient assisted living environments to maintain their privacy and security. The datasets will be used to evaluate the performance of algorithms that can learn to automatically remove any private information that the residents don't want other people to have, especially on video data. DS1, DS2, DS6, DS8
16 Neural Information Processing Institute, University of Ulm, Germany Yan Zhang It is important to propose an automatic and unobstructive method to recognize abnormal behaviors of elderly people. In our work, we proposed a novel unsupervised online learning algorithm to group the body skeletons overtime and learn the long-term dynamics gradually, so that static abnormal behaviors and abnormal movements can be detected. This algorithm first determine the number of clusters using an novel graph spectrum algorithm; then the codebook will update in an online manner as perceiving data. In addition, the transition probability between keywords in codebook is updated as well. The skeleton extraction and analysis algorithm are implemented in GPU. As tested on other dataset, it can run in real-time.  DS1, DS2, DS4, DS6, DS8
17 Dept of Computer Science and Engg.
Indian Institute of Technology Patna
Jimson Mathew
Associate Professor and Head
Our project aim at robust activity recognition using data from multiple sensors. We aim to study the effect of fusing multi-sensor data in accuracy of activity recognition system. RGB-D data will help our model tobe keen attention on the person in sight from the background and extractmore details than RGB data. Motion, physiological and plug sensor data
will be fused with RGB-D data for improving recognizing the activity.
Features extracted from this dataset will be used to trained our machinelearning model. The model will be able to predict abnormal activities and alert as necessary.
DS1,DS4, DS6, DS8, DS9,
DS10, DS11, DS12
18 Department of Computer Engineering, University of Engineering and Technology, Taxila, Pakistan Dr. Muhammad Awais Azam / Assistant Professor This research work aims to provide an assistant and independent living for human beings based on human behavior modeling and activity interpretation in their living and working environments. The motivation is to formulate a universal framework for continuous monitoring of human behavior, which is based on recognition of activities of daily living, detection of changes in those activities and ultimately the detection of abnormal and/or unforeseen activities. This dataset will be used to detect the activities by monitoring physical, physiological, visual and contextual parameters related to human beings using multiple sensor modalities. Features extracted from this dataset will be used to train different machine learning classifiers, which will be able to detect, recognize and predict unusual behavior of a person. Detection of unforeseen and abnormal human behavior is useful for the safety and healthcare of human beings and necessary aid and/or guidance can be provided to the persons when needed. DS1,DS2,DS4,DS6,DS8,DS9,
19 Semnan university, Iran Dr. Hadi Soltanizadeh /Assistant Professor Early detection of Alzheimer disease in old people using behavioral features such as speech, gait and
daily activity. We are working on new ICT technology to recognize Alzheimer in old people in the early
stage of disease. We have been working on speech, gait and daily activity to detect Alzheimer in old
people and also to classify old people to different groups such as normal, MCI and AD.
DS1, DS3, DS5, DS6, DS7, DS8

© 2011 Dem@Care | Centre for Research and Technology Hellas :: Information Technologies Institute