05.06.2013 Talk from Thomas Ploetz "Activity Recognition for Real World Health Scenarios"

Title: Activity Recognition for Real World Health Scenarios

When: Wednesday, 05.06.2013, 15:00h

Activity recognition aims for automatically detecting and analysing (largely human) behaviour thereby focusing on movements, the whereabouts of the person(s) of interest, and their interaction with the environment. With the raise of ubiquitous / pervasive / wearable computing this broader concept of "context aware computing" has become very popular in the recent past.

Context aware computing, and especially activity recognition, relies on sophisticated data analysis techniques. This largely corresponds to robust and reliable machine learning methods. However, in order to successfully tackle real-world problems these methods need to be wisely chosen and typically require domain adaptation on a per use-case basis.


In this seminar I will give an overview of research activities of the Digital Interaction group at Culture Lab, Newcastle University, that are centred on context awareness for real-life applications. I will present several concrete use-cases of health-related applications of activity recognition and will illustrate in detail how in practice opportunistic sensing and bespoke sensor data analysis techniques are integrated to have an impact on people's life.

Thomas Ploetz is a Lecturer (Assist. Prof.) in "Context Aware Computing" at Newcastle University (Newcastle upon Tyne, UK). 


His research agenda is centred on "Computational Behaviour Analysis", which basically means that he is building computational (that is statistical) models that describe and will help in assessing behaviour. The basis for this is the analysis of behavioural data that is captured in a pretty opportunistic way utilising a variety of sensing modalities, most notably pervasive/ubiquitous sensors (e.g., accelerometers, RFID, environmental sensors), cameras, or microphones. The modelling itself is agnostic in terms of the actual choice of sensing modalities as long as the relevant information for behaviour analysis is captured. Behaviour data are sequential by definition. Consequently, related modelling techniques are focused on sequential models (for example of Markovian type). He is especially interested in quantitative assessments of human behaviour, which is of value for, for example, skill assessment. Thomas' day-to-day work can probably best be summarised as applied machine!


The central theme of Thomas' research is to develop techniques and systems that actually have an impact on people's life. Therefore, his research is almost always connected to some practical application (in contrast to purely theoretical work) and he is keen on deploying systems he develops in the "wild", i.e., in real-world settings. The most prominent domain for this kind of work is health, where he is working on computational assessments of behavioural phenotypes of, for example, Parkinson's, Dementia, or Autism.