�ݺ�ߣ

Processing Mobility Traces for Activity
Recognition in Smart Cities
Date: November 2017
Contact information
Tampere University of Technology,
FAST Laboratory,
P.O. Box 600,
FIN-33101 Tampere,
Finland
Email: fast@tut.fi
www.tut.fi/fast
Conference:11th International Workshop
on Service-Oriented Cyber-Physical
Systems in Converging Networked
Environments (SOCNE). Beijing, China,
October 29 - November 1, 2017
Title of the paper: Processing Mobility
Traces for Activity Recognition in Smart
Cities
Authors: Arsalan Shah, Petr Belyaev,
Borja Ramis Ferrer, Wael M. Mohammed,
Jose L. Martinez Lastra
If you would like to receive a reprint of
the original paper, please contact us
31.10.2017
Processing Mobility Traces for Activity Recognition in
Smart Cities
1

Processing Mobility Traces for
Activity Recognition in Smart Cities
Arsalan Shah, Petr Belyaev, Borja Ramis Ferrer, Wael M. Mohammed, Jose L.
Martinez Lastra
{syed.a.shah, petr.belyaev, borja.ramisferrer, wael.mohammed, jose.lastra}@tut.fi
Tampere University of Technology, Tampere, Finland
11th International Workshop on Service-Oriented Cyber-Physical Systems in Converging
Networked Environments (SOCNE2017) in the 43rd Annual Conference of the IEEE
Industrial Electronics Society (IECON2017)
1st November 2017, China National Convention Center, Beijing, China

Outline
Smart Cities
3
• Introduction
• Motivation
• Objectives
• Problem definition
• Inputs and Outputs for ANFIS System
• The Approach (Activity Recognition) (1/3)
• Implementation
• Adaptive Neuro-Fuzzy Inference System
• Results
• Conclusion
31.10.2017

Smart Cities
4
Introduction
• Human mobility modelling has emerged as an important
research area over the past years
– From smart transportation services to reliable
recommendations systems
• Activity recognition emerges as a vital initial step towards
building better and accurate human mobility models
• Analyze human mobility data i.e. GPS traces
– Identify activities from the traces using mobility history
31.10.2017

Smart Cities
5
Motivation
• Vast amounts of mobility data generated
• Mobility traces of the inhabitants enhance transportation
services of cities
– Provide better services to users
• Offer the possibility to make existing apps “smarter”
31.10.2017

Smart Cities
6
Objectives
• To identify and label activities a user has performed by
analyzing the past and current mobility traces
• Propose a framework on how to handle sensitive mobility
data and utilize it for activity recognition
31.10.2017

Smart Cities
7
Problem Definition (1/2)
Mon – Fri
08:00 – 16:00
Alternate Days
16:30 – 17:30
Weekly
19:00 – 20:30
Weekends
18:00 – 19:00 Everyday
22:00 – 06:30
31.10.2017

Smart Cities
8
Problem Definition (2/2)
Mon – Fri
08:00 – 16:00
Alternate Days
16:30 – 17:30
Weekly
19:00 – 20:30
Weekends
18:00 – 19:00
Everyday
22:00 – 06:30
31.10.2017

Inputs and Outputs for ANFIS System
• System input
– Start time of an activity
– Time spent doing an activity
• System outputs (activities)
– Work
– Leisure & chores
– Eating
– At home
Smart Cities
931/10/2017

Smart Cities
10
The Approach (Activity Recognition) (1/3)
31.10.2017
Stay points
extraction
(POIs)
Processing
POIs data
ANFIS
model
Activity
label as
output
1 2 3 4
Use labelled data to train
and obtain ANFIS model

Smart Cities
11
1. Stay points extraction (Points of Interest)
i. Time spent at a location
ii. Distance between the farthest points
2. Processing each POI to obtain
i. Radius for each POI
ii. Recalculation of radius with Chebyshev inequality
iii. Merging similar POIs
iv. Time at which the person reached a POI
v. Time spent at each POI
31/10/2017

Smart Cities
12
3. Input variables to already generated ANFIS model
i. Starting time
ii. Time spent
4. Apply thresholding to identify activity
31/10/2017

Implementation (1/2)
Smart Cities
1331.10.2017

Implementation (2/2)
Smart Cities
14
what can I do?
31.10.2017

Adaptive Neuro-Fuzzy Inference System (1/5)
• ANFIS (MATLAB) toolbox
• Training of fuzzy model
– Data gathered from researchers through google maps
– Researchers asked to label the extracted stay points
• Labelled data and input parameters used to train the fuzzy
model
• Using intuitive rules to make an alternative fuzzy model
• The Fuzzy model used in step 3 is obtained
Smart Cities
1531.10.2017
Stay points
extraction
(POIs)
Processing
POIs data
ANFIS
model
Activity
label as
output
1 2 3 4
Use labelled data to train
and obtain ANFIS model

• General structure
Smart Cities
1631.10.2017

• Rules
Smart Cities
1731.10.2017

• Model surfaces (Activity 1 & Activity 2)
Smart Cities
1831/10/2017

• Model surfaces (Activity 3 & Activity 4)
Smart Cities
1931/10/2017

Results (1/2)
• Visualized using google maps API
Smart Cities
2031/10/2017

Results (2/2)
Smart Cities
2131/10/2017

Conclusions
• Fuzzy logic for recognition of activities is similar to
human decision making
• An approach for recognizing activities from the GPS
traces is presented
• Start with a general model when no data is available
– Learning slowly from the user, based on the data
Smart Cities
2231/10/2017

31.10.2017
Smart Cities
23
• The project leading to this paper has received funding
from the European Union’s Horizon 2020 research and
innovation programme under grant agreement n°
644429 correspondent to the project shortly entitled
MUSA, Multi-cloud Secure Applications
Acknowledgement

THANK YOU!
Any questions?
youtube.com/user/fastlaboratory
facebook.com/fast.laboratory
slideshare.net/fastlaboratory
twitter.com/FAST_Lab
31.10.2017
Smart Cities
24

�ݺ�ߣ

Processing Mobility Traces for Activity Recognition in Smart Cities

More Related Content

Processing Mobility Traces for Activity Recognition in Smart Cities