Unobtrusive Video Monitoring Analysis

People are naturally anxious about being videoed in their own homes so the approach that we have taken is based on extracting a '3D skeleton' description of the observed person, and then immediately discarding the original video. This process avoids recording the appearance of the monitored person (who were shown the skeletons), thus preserving privacy while still acquiring rich data for several different types of monitoring.

The team investigated both short and long term monitoring.

Two types of short-term monitoring were investigated:

1. Assessing if a person has entered a room but not exited, which is particularly relevant to bathrooms, as many accidents occur because of wet slick floors.
2. Assessing if a person has recently stopped moving (as most people make minor body movements every few seconds, even when sleeping).

Long-term (e.g. weeks or months) monitoring investigations were:

1. How arm speed and distance reached varies over time as a person eats (which may be an indicator of increasing frailty).
2. Counting of mouthfuls of food eaten (which might be an indicator of malnutrition).
3. How walking speeds and gait varies over time as a person walks down their hallway (which may be an indicator of short- or long-term declining health or improving fitness after e.g. a hip replacement).
4. How general activity levels (i.e. general body motion) change over time.
5. How general activity levels change after some intervention, such as fasting, change of diet, new hobby.

This research has resulted in the advanced training of two MSc, two PhD and one post-doctoral researchers. The results have been presented in seven journal papers and seven peer-reviewed full conference papers so far.