WEBVTT

1
00:00:00.500 --> 00:00:02.000
Hello everybody!

2
00:00:02.000 --> 00:00:06.000
In this presentation, we will talk about our research illustrated in the paper

3
00:00:06.000 --> 00:00:13.000
“Exploring Mobility Behavior Around Ambient Displays Using Clusters of Multi-dimensional Walking Trajectories”.

4
00:00:14.300 --> 00:00:17.300
Our work concerns the field of ambient display research

5
00:00:17.300 --> 00:00:21.300
where studies have historically been evolving from not including any evaluations at all,

6
00:00:21.300 --> 00:00:25.500
in anecdotal examples (for instance, in scientific laboratories)

7
00:00:25.500 --> 00:00:31.300
to, nowadays, sophisticated field deployments using modern sensors such as Microsoft Kinect cameras.

8
00:00:31.300 --> 00:00:36.000
Today, researchers are trying to find out better how people naturally behave around ambient displays.

9
00:00:36.000 --> 00:00:40.900
However, with this development came problems that brought new challenges as, for example,

10
00:00:40.900 --> 00:00:44.600
researchers are now facing vast amounts of data that need to be analyzed.

11
00:00:44.600 --> 00:00:50.700
Especially the analysis of skeletal data drawn from camera sensors lacks higher levels of automation.

12
00:00:50.700 --> 00:00:57.700
This led to the situation that mobility behavior captured with such cameras still cannot be readily interpreted.

13
00:00:57.700 --> 00:01:01.000
Our research aims to advance on this issue.

14
00:01:01.500 --> 00:01:07.000
Mobility behavior has long been investigated using one- or multiple-dimensional walking trajectories.

15
00:01:07.000 --> 00:01:13.200
To analyze them, time series clustering has been introduced as one of the most important and useful tools to do so.

16
00:01:13.200 --> 00:01:18.400
We draw attention to agglomerative hierarchical clustering and dynamic time warping.

17
00:01:18.400 --> 00:01:23.000
To our knowledge, both algorithms have not found application in our field so far.

18
00:01:23.000 --> 00:01:27.000
Our contribution is, therefore, more at the methodological level.

19
00:01:27.500 --> 00:01:33.300
For demonstrating purposes, we analyzed an existing data set obtained from a real-world field deployment study

20
00:01:33.300 --> 00:01:38.300
and leveraged different visualizations to evaluate our approach’s capabilities.

21
00:01:38.300 --> 00:01:42.500
Now, by considering the clustering algorithm’s resulting dendrogram,

22
00:01:42.500 --> 00:01:47.300
we knew that the data set may contain many walking trajectories similar in shape.

23
00:01:47.300 --> 00:01:51.300
We then experimented with further illustrations at different cutoff levels

24
00:01:51.300 --> 00:01:54.700
to unveil the underlying patterns and, at the same time,

25
00:01:54.700 --> 00:01:59.600
double-checked our findings with the original study where we were taking the data set from.

26
00:02:00.000 --> 00:02:05.000
It became evident that our approach was indeed able to correctly identify and group together

27
00:02:05.000 --> 00:02:07.900
similar two-dimensional walking trajectories.

28
00:02:08.300 --> 00:02:10.300
How is this useful, you might ask?

29
00:02:10.300 --> 00:02:14.500
So, what we can take away from our research is that there are now practical tools

30
00:02:14.500 --> 00:02:19.400
to more rapidly extract mobility behavior underlying a skeletal data set and,

31
00:02:19.400 --> 00:02:24.000
simultaneously, to notably reduce manual efforts throughout this process.

32
00:02:24.300 --> 00:02:27.000
Our work is, however, not without limitations

33
00:02:27.000 --> 00:02:31.000
and there is more work to be done to fully automate skeletal data analysis.

34
00:02:31.400 --> 00:02:38.000
In the near future, we will concentrate on tasks such as optimizing the computational complexity of the algorithms

35
00:02:38.000 --> 00:02:43.000
and extending the existing approach to, for example, suggest the optimal number of clusters.

36
00:02:43.500 --> 00:02:46.500
With this brief outlook, our presentation ends.

37
00:02:46.500 --> 00:02:50.000
For more information, please consider having a look at our publication

38
00:02:50.000 --> 00:02:53.500
and please do not hesitate to contact us in case of any questions.

39
00:02:53.500 --> 00:02:55.000
Thank you!