Transcript ppt

Mapcube to Understand
Traffic Patterns
Shashi Shekhar
Computer Science Department
University of Minnesota
[email protected]
(612) 624-8307
http://www.cs.umn.edu/~shekhar
http://www.cs.umn.edu/research/shashi-group/
Motivation for Traffic Visualization

Transportation Manager



Traffic Engineering





Where are the congestion (in time and space)?
Which of these recurrent congestion?
Which loop detection are not working properly?
How congestion start and spread?
Traveler, Commuter




How the freeway system performed yesterday?
Which locations are worst performers?
What is the travel time on a route?
Will I make to destination in time for a meeting?
Where are the incident and events?
Planner and Research


How much can information technique to reduce congestion?
What is an appropriate ramp meter strategy given specific evolution of
congestion phenomenon?
Contributions

Transportation Domain



Allow intuitive browsing of loop detector data
Highlight patterns in data for further study
Computer Science


Mapcube - Organize visualization using a dimension lattice
Visual data mining




Hotspots, clustering (slides 9, 10 & 11)
Discontinuity, Sharp Gradients, Discontinuities (slides 7 & 11)
Co-locations, co-occurrences
Location classification and predication (slide 13)
Map of Station in Mpls
Dimensions

Available
•
•
•
•

TTD : Time of Day
TDW : Day of Week
TMY : Month of Year
S : Station, Highway, All Stations
Others
• Scale, Weather, Seasons, Event types, …
Mapcube :
Which Subset of Dimensions ?
TTDTDWTMYS
TTDTDWS
TTDTDW
TDW
S
STTD
TTD
TDW
S
Next Project
Singleton Subset : TTD
Configuration:  X-axis: time of day; Y-axis: Volume
 For station sid 138, sid 139, sid 140, on 1/12/1997
Trends:
 Station sid 139: rush hour all day long
 Station sid 139 is an S-outlier
Singleton Subset: TDW

Configuration: X axis: Day of week; Y axis: Avg. volume.
For stations 4, 8, 577
Avg. volume for Jan 1997
Trends:
Friday is the busiest day of week
Tuesday is the second busiest day of week
Singleton Subset: S
Configuration:  X-axis: I-35W South; Y-axis: Avg. traffic volume
 Avg. traffic volume for January 1997
Trends?:

High avg. traffic volume from Franklin Ave to Nicollet Ave

Two outliers: 35W/26S(sid 576) and 35W/TH55S(sid 585)
Dimension Pair: TTD-TDW
Configuration:
Trends:



X-axis: time of date; Y-axis: day of Week

f(x,y): Avg. volume over all stations for Jan 1997, except
Jan 1, 1997

Evening rush hour broader than morning rush hour
Rush hour starts early on Friday.
Wednesday - narrower evening rush hour
Dimension Pair: S-TTD
Configuration:



X-axis: Time of Day
Y-axis: Route
f(x,y): Avg. volume over all stations for
1/15, 1997
Trends:



3-Clusters
• North section:Evening rush hour
• Downtown area: All day rush hour
• South section:Morning rush hour
Spatial Outliers, Discontintuities
• station ranked 9th
• Time: 2:35pm
Missing Data
Dimension Pair: TDW-S
Configuration:
Trends:




X-axis: stations; Y-axis: day of week

f(x,y): Avg. volume over all stations for Jan-Mar 1997
Busiest segment of I-35 SW is b/w Downtown MPLS & I-62
Saturday has more traffic than Sunday
Outliers – Route branch
Post Processing of cluster patterns
Clustering Based Classification:
 Class 1: Stations with Morning Rush Hour
 Class 2: Stations Evening Rush Hour
 Class 3: Stations with Morning + Evening Rush Hour
Size 4 Subset: TTDTDWTMYS(Album)
Configuration:
Trends:
 Outer: X-axis (month of year); Y-axis (route)
 Inner: X-axis (time of day); Y-axis (day of week)
 Morning rush hour: I-94 East longer than I-35 W North
 Evening rush hour: I-35W North longer than I-94 East
 Evening rush hour on I-94 East: Jan longer than Feb
Triplet: TTDTDWS: Compare Traffic Videos
Configuration: Traffic volume on Jan 9 (Th) and 10 (F), 1997
Trends:
 Evening rush hour starts earlier on Friday
 Congested segments: I-35W (downtown Mpls – I-62);
I-94 (Mpls – St. Paul); I-494 ( intersection I-35W)
Data Fusion levels and Mapcube


Different Sub-cubes help with different data fusion levels
Level 0: Single Sensor


Level 1: Correlating Multiple Sensors



Local weather as a function of time
Map of spatial variation in weather
Space-time plot for a route for a day
Level 2: Interpret, Aggregate



Detect spatial discontinuities, spatial outliers
Group sensors with similar weather measurements
Group timeslots with similar weather measurements
Spatial Data Mining, SDBMS

Historical Examples



London Cholera (1854)
Dental health in Colorado
Current Examples





Environmental justice
Crime mapping - hot spots (NIJ)
Cancer clusters (CDC)
Habitat location prediction (Ecology)
Site selection, assest tracking, spatial outliers