The Crosswatch Traffic Intersection
Analyzer: a Roadmap for the Future


       James Coughlan, Ph.D.
Traffic intersections: difficult
    for many visually impaired
              travelers
How to align oneself
 properly to crosswalk?
What is the layout of the
 intersection?
How to remain inside
 crosswalk while
 walking?
When is it safe to cross?
How to avoid moving               2
 vehicles?
Past work
“Crosswatch” system (Ivanchenko et al.,
ICCHP 2010; etc.) – computer vision
smartphone app that performs the
following functions in real time for visually
impaired user:
•find crosswalk markings (e.g., zebra
stripes), determine user’s orientation
relative to crosswalk “corridor”
•find Walk light and announce if it’s on
                                                3
Crosswatch (past work)



Two kinds of misalignment relative to
crosswalk corridor: (1) translated to the
left; (2) direction error.

Crosswatch can calculate both kinds of
misalignment.
                                            4
Crosswatch (past work)




How Crosswatch finds crosswalk and
calculates alignment                 5
Crosswatch (past work)




Crosswalk scene   Zoomed-in
                  region showing
                  Walk light       6
Advantages of
   Crosswatch approach
Information provided to user in real time

System requires no new infrastructure
since it analyzes the visual features that
are already there



                                             7
Related work
Aranda & Mares, ICCHP 2004: computer
vision for Walk light detection

Zebralocalizer (Ahmetovic, Bernareggi &
Mascetti, 2011): smartphone computer
vision system

Bohonos et al., ICCHP 2008: Bluetooth
beaconing at traffic intersections
                                          8
Next steps
Travelers need a broad range of
information about intersections:
•What type of intersection, features in it
•Where are the crosswalk and other
important features
•When is it time to cross?



                                             9
“What” information
What is the intersection
layout? Four-way, three-
way, etc.?
Are there stop signs?
Signal lights? Walk light?
Is there a pedestrian
walk button?
Is there a median strip
somewhere in the
crosswalk?
                              10
“Where” information
Where are all the crosswalks? How am I
oriented relative to the one I want?

Where is the pedestrian walk button relative to
where I am standing?

How long is the crosswalk? How far is it to the
median strip?

                                            11
“When” information
When is it time to cross, according to Walk light
(or traffic lights if there is no Walk light)?

Note: this is not necessarily the same as when
it is safe to cross!




                                             12
How to get What, Where,
     When information?
Realization: it is impractical to expect computer
vision to figure out all this information on its
own!

Intersections, and the features in them, are
extremely variable, and not always easy to see
(even for people with normal vision)


                                             13
Walk buttons come in
many shapes and sizes




                        14
Find the walk button!




                        15
Median strips
Without high-
resolution depth
information to show
the elevated median
strip, it is hard to
find reliably


                             16
Complicated intersections
Where a visually impaired traveler may
need the most guidance… computer
vision may have the most difficulties!




                                         17
Solution: GIS
Use computer vision where it is
necessary:
(b)find user’s location and orientation
relative to intersection features
(b) detect signal lights in real time (e.g.,
Walk light)

Augment with GIS (Geographic
Information Systems)                           18
How this works
GPS tells you which intersection you’re at
Smartphone compass and accelerometer
indicate direction you’re pointing in
Computer vision analyzes visible features
in scene and determines (x,y) GPS
coordinates (to within about 1 meter)
Use GIS to find information relevant to
current location (look up online, or
download all information for a city or
region onto smartphone)                  19
How this works (continued)
GIS has data for
each intersection,
including layout of
all crosswalks, list
of important
features and
landmarks and their
locations
                        20
Much to do
Where to find this GIS information?

Some available through municipal/transit
agencies, Google Maps

But much of it may require crowd-sourcing:
volunteers will contribute information about
intersections in their neighborhood to
online GIS database                       21
Much to do (continued)
Develop Crosswatch algorithms for
estimating high-resolution GPS
coordinates

Appropriate Crosswatch user interface: for
instance, how to help user pan left to right
to capture entire intersection scene

User surveys (what should Crosswatch do
                                      22

and not do?) and testing
Conclusion
Crosswatch needs to expand its
functionality to provide extensive “what”,
“where” and “when” information

Use computer vision where it is needed
and rely on GIS to provide other
information (data infrastructure)

Future idea: use GIS to help with route
planning that is tailored to needs of traveler
                                            23
Thanks to…
Second author: Dr. Huiying Shen (Smith-
  Kettlewell)

Collaborators: Dr. Megan Lawrence
  (Smith-Kettlewell), Dr. Volodymyr
  Ivanchenko (Amazon Lab 126)

Funding from NIH and NIDRR



                                          24

The Crosswatch Traffic Intersection Analyzer: A Roadmap for the Future - Coughlan, James (s)

  • 1.
    The Crosswatch TrafficIntersection Analyzer: a Roadmap for the Future James Coughlan, Ph.D.
  • 2.
    Traffic intersections: difficult for many visually impaired travelers How to align oneself properly to crosswalk? What is the layout of the intersection? How to remain inside crosswalk while walking? When is it safe to cross? How to avoid moving 2 vehicles?
  • 3.
    Past work “Crosswatch” system(Ivanchenko et al., ICCHP 2010; etc.) – computer vision smartphone app that performs the following functions in real time for visually impaired user: •find crosswalk markings (e.g., zebra stripes), determine user’s orientation relative to crosswalk “corridor” •find Walk light and announce if it’s on 3
  • 4.
    Crosswatch (past work) Twokinds of misalignment relative to crosswalk corridor: (1) translated to the left; (2) direction error. Crosswatch can calculate both kinds of misalignment. 4
  • 5.
    Crosswatch (past work) HowCrosswatch finds crosswalk and calculates alignment 5
  • 6.
    Crosswatch (past work) Crosswalkscene Zoomed-in region showing Walk light 6
  • 7.
    Advantages of Crosswatch approach Information provided to user in real time System requires no new infrastructure since it analyzes the visual features that are already there 7
  • 8.
    Related work Aranda &Mares, ICCHP 2004: computer vision for Walk light detection Zebralocalizer (Ahmetovic, Bernareggi & Mascetti, 2011): smartphone computer vision system Bohonos et al., ICCHP 2008: Bluetooth beaconing at traffic intersections 8
  • 9.
    Next steps Travelers needa broad range of information about intersections: •What type of intersection, features in it •Where are the crosswalk and other important features •When is it time to cross? 9
  • 10.
    “What” information What isthe intersection layout? Four-way, three- way, etc.? Are there stop signs? Signal lights? Walk light? Is there a pedestrian walk button? Is there a median strip somewhere in the crosswalk? 10
  • 11.
    “Where” information Where areall the crosswalks? How am I oriented relative to the one I want? Where is the pedestrian walk button relative to where I am standing? How long is the crosswalk? How far is it to the median strip? 11
  • 12.
    “When” information When isit time to cross, according to Walk light (or traffic lights if there is no Walk light)? Note: this is not necessarily the same as when it is safe to cross! 12
  • 13.
    How to getWhat, Where, When information? Realization: it is impractical to expect computer vision to figure out all this information on its own! Intersections, and the features in them, are extremely variable, and not always easy to see (even for people with normal vision) 13
  • 14.
    Walk buttons comein many shapes and sizes 14
  • 15.
    Find the walkbutton! 15
  • 16.
    Median strips Without high- resolutiondepth information to show the elevated median strip, it is hard to find reliably 16
  • 17.
    Complicated intersections Where avisually impaired traveler may need the most guidance… computer vision may have the most difficulties! 17
  • 18.
    Solution: GIS Use computervision where it is necessary: (b)find user’s location and orientation relative to intersection features (b) detect signal lights in real time (e.g., Walk light) Augment with GIS (Geographic Information Systems) 18
  • 19.
    How this works GPStells you which intersection you’re at Smartphone compass and accelerometer indicate direction you’re pointing in Computer vision analyzes visible features in scene and determines (x,y) GPS coordinates (to within about 1 meter) Use GIS to find information relevant to current location (look up online, or download all information for a city or region onto smartphone) 19
  • 20.
    How this works(continued) GIS has data for each intersection, including layout of all crosswalks, list of important features and landmarks and their locations 20
  • 21.
    Much to do Whereto find this GIS information? Some available through municipal/transit agencies, Google Maps But much of it may require crowd-sourcing: volunteers will contribute information about intersections in their neighborhood to online GIS database 21
  • 22.
    Much to do(continued) Develop Crosswatch algorithms for estimating high-resolution GPS coordinates Appropriate Crosswatch user interface: for instance, how to help user pan left to right to capture entire intersection scene User surveys (what should Crosswatch do 22 and not do?) and testing
  • 23.
    Conclusion Crosswatch needs toexpand its functionality to provide extensive “what”, “where” and “when” information Use computer vision where it is needed and rely on GIS to provide other information (data infrastructure) Future idea: use GIS to help with route planning that is tailored to needs of traveler 23
  • 24.
    Thanks to… Second author:Dr. Huiying Shen (Smith- Kettlewell) Collaborators: Dr. Megan Lawrence (Smith-Kettlewell), Dr. Volodymyr Ivanchenko (Amazon Lab 126) Funding from NIH and NIDRR 24