IMMED
Indexing Multimedia Data from Wearable Sensors for diagnostics and treatment of Dementia
Problem description
With the ageing of population, dementia cases consequently increases in Europe. An early diagnosis prevent insecurity and health worsening in aged people living at home. Generally, dementia is diagnosed in gathering clues of pathological changes in people life. To assess those changes, physicians use neurological examination, neuropsychological testing, and brain imaging technologies. Finally, diagnosis of possible dementia is asserted by comparing evidences, added to an autonomy decline. This autonomy decline is frequently assessed by Activities of Daily Living interviews in which impairments are known to be related to cognitive decline caused by dementia.
Measuring autonomy decline is thus part of the diagnosis process, but it depends on subjective tools and on the patients ability to clearly analyze situations.
Our intention in this project is to develop a new system to assess cognitive decline in the patients’ life, and assist physician to make a diagnostic. This would help in people rehabilitation, by offering strategies to maintain their autonomy at home.
Objectives
We develop a wearable video device that allows a capture of the patients’ daily activities. By the use of this new device, we collect video data that can be analyzed to extract meaningful events occurring in patient’s everyday life. To assist physician in the analysis of video data, we create a video indexing assistance. This indexing process uses video and audio media to automatically guide the navigation in data flow straight up to the meaningful recorded situations.
The Samova team is particulary involved in audio indexing and cross media fusion. An objective of audio indexing is to detect some Activity of Daily Living of the patient wearing the video device, by automatically analyze the audio stream extracted from the video.
Methodology
The sounds events recognition in a noisy environment is a problem very hard to solve. The datas of the project are moreover particularly difficult to analyze, due to the acoustic variability of the places and the objects in which the patient move or interact. Therefore, we work actually on on sound events recognition system without machine learning. The basics of this system is to detect sound events that are particularly useful to give an indication on the activity of the patients.
Collaborations
People involved in SAMOVA team
- Julien Pinquier (scientific coordinator)
- Patrice Guyot
- Régine André-Obrecht
- Philippe Joly
Support
- ANR Blanc (2009 call)
Scheduled
- Start time : 1st october 2009
- End time : 30th september 2012