Urban Computing with Taxicabs

Report
Urban Computing with Taxicabs
TMSG
Cappuccino
Oct. 12th, 2011
About the Authors..
• Yu Zheng, Yanchi Liu, Jing Yuan, Xing Xie
from M$RA
Yu Zheng
•Ubicomp - Ubiquitious computing group
•WSM - Web Search & Mining Group
Yu Zheng’s weibo : http://www.weibo.com/msyuzheng
About the Authors..
• Yu Zheng, Yanchi Liu, Jing Yuan, Xing Xie
from M$RA
• Yanchi Liu, also from University of Science and
Technology Beijing
• Paper published by ACM on Sept. 2011 ( ? )
Agenda
•
•
•
•
•
•
•
•
About the authors(finished)
Introduction
Overview
Evaluation Settings
Evaluation Results
Related Works
Conclusion
Comments
Introduction
• GPS equips are generally used nowadays.
• GPS-equipped taxicabs can be viewed as
ubiquitious mobile sensiors, gathering datas of
traffic flows.
Introduction
• This paper’s goal is to detect the flawed and
less effective urban planning in a city
according to the GPS trojectories of taxicabs.
• Saying Beijing has 35 million personal trips per day
created by various kinds of vehicles and that 1.44
million personal trips are generated by taxi. Thus, the
percentage of taxi trips from total trips are 4.2%.
The author believed 4.2% is a significant
sample reflecting to urban traffic flow.
Overview
• Let’s introduce the variables one at a time..
Overview
Overview
Overview
• Consisted of two major components:
Modeling citywide traffic
Detecting flawed planning
Overview
• Disjoint regions using
major roads.
Overview
• Building region matrix: This process can be
break down into 3
steps...
• 1. Temporal partition
• Weekday/weekend
• Rush hours/un-rush hours
Overview
• 1. Temporal partition
Overview
• 2. Transition construction
• Pick out effective(occupied) taxi trajectories
• Construct transitions between two regions
according to Def. 3
Overview
• 2. Transition construction
• Note that a trajectory discontinuously
traversing two regions still formulate a
transition between two regions.
Overview
• 2. Transition construction
Overview
• 2. Transition construction
Black point represents
a region pair, while blue
and red points are the
projections of these region pairs on XZ and YZ
spaces.
Overview
• 2. Transition construction
Most taxis intend to travel through a shortcut
instead of the roundabout route if the shortcut is
effective. On the contrary, if most taxis pass
additional regions, that means the route directly
connecting two places is not very effective.
Overview
• 3. Build region matrix
Overview
• Detecting flawed planning
Skyline detection
Overview
• In this paper, which means there’s no region
pair a(p, q) having a lower speed and bigger
theta than those belongs to skyline.
Overview
• The detected skyline is comprised of 3 kinds of
region pairs.
1. Very small E(V) and theta, which means the
two regions are connected with some direct
routes while the capacity of these routes are not
sufficient.
Overview
2. A region pair with a small E(V) and big theta,
which means people have to take detours and
also suffer from a very slow speed. Worse case.
3. A region pair with a big E(V) and big theta.
Meaning that the travel speed is fast but far, still
has flaws.
Overview
Examples of 1, 2, and 3.
Overview
• Pattern mining from skylines(2 steps)
1. Formulating skyline graphs
2. Mining frequent sub-graph patterns
Overview
1. Formulating skyline graphs
connect two consecutive slots
if they are spatially close to each
other.
There could be multiple
isolated graphs pertaining
to a day.
Overview
1. Formulating skyline graphs
Overview
2. Mining frequent sub-graph patterns
Mining the association rules.
The mined association rules can consist of over 2 patterns. E.g. g1,
g2 => g3. Also, these association rules may NOT be geospatially
close to each other.
Evaluation Settings
• Taxi trajectories
Evaluation Settings
• Map data
Road network of Beijing, consist of 106,579 road
vertices and 141,380 road segments.
Picked out 25,262 road segments w/ leveling from
0 to 6. Use only the 0 to 2 level. (0 is highest
representing highways)
Create 444 regions in result.
Evaluation Settings
• Verify the detected flaws in the following 2
ways:
1. Verify the urban plannings that had been
implemented between the times of the two
datasets.
2. Check if some flaws that have been detected in
both datasets by our methods embodied in the
future urban planning of Beijing.
Evaluation Results
Evaluation Results
• Sum up that the traffic conditions in Beijing
become worse in 2010 than 2009.
Evaluation Results
• Taxi drivers took fewer passengers than b4.
• The average speed dropped between the two
years.
• In short, the traffic condition became worse.
Evaluation Results
• Most regions becomes
shallower in 2010 than
2009, especially in hot
areas.
• The travel speed of
taxis in these regions
decreased.
Evaluation Results
• By looking at the results, we observe two
aspects:
1. Some flawed planning occurring in 2009
disappeared in 2010.
2. The number of regions having defects
increased in 2010 beyond 2009 and some
flaws occuring in 2009 still exist.
Evaluation Results
• Some flawed planning occurring in 2009
disappeared in 2010.
Evaluation Results
• Some flawed planning that still exists in 2010.
The planning of these two
subway lines denotes that
the urban planner has
recognized the problem
existing in the regions,
justifying the validity of the
results generated using our
method.
Evaluation Results
• Can also found association rules between the
detected patterns.
Support = 0.05
Confidence = 0.7
Evaluation Results
• We will show more interesting results and a
live demo during the presentation at the
conference.
Related works
• Mining taxi trajectories
Effectiveness, like what I’ve presented last time.
Destination prediction, like what Danny presented
last time.
• Urban computing
Most papers do their researchs on social computing
and some user interaction stuff while this paper
explore the urban computing from the perspective
of urban planning.
Conclusion
• This paper detect the flaws in the existing
urban planning of a city using the GPS
trajactories of taxis traveling in the urban
areas.
Conclusion
• Future plans:
1.Studying the geographic features of a region, such
as the road segments and POI.
2. The purpose of people’s travel.
Comments
•
•
•
•
Easy reading
Interesting and well-applied on a useful case
Chart arrangement thing again.
Like reading a novel or comic, no highlights.
Urban Computing with Taxicabs
The End 

similar documents