Counting Pedestrians in Urban Environments

1
Counting Pedestrians in Urban Environments

• Comparison of Manual and Automated Counting
  Methods

2
Outline

• Why count pedestrians?
• Counting methods
• Our study
• Further needs / research

Dan Burden
3
Why count pedestrians?
4
Why count pedestrians?

• Monitor pedestrian risk
• Test effectiveness of safety countermeasures
• Target infrastructure investments
• Understand correlates of walking
• We do it for all the other modes, so why not for
  peds?

5
Study questions

• What is the relative accuracy of pedestrian
  counting methods?
• What are the strengths and weaknesses of each
  method?

6
Methods

• Manual with clipboard
• Manual with clickers
• Manual with video review
• Automated methods
• Active and passive infrared
• Laser scanners
• Piezo pads
• Computer vision

7
Passive Infrared (Dual sensor)

• Pros
• Easy mounting
• Relatively affordable
• No liability issues
• Easy data download
• Cons
• Trouble counting adjacent peds

8
Methods

• Two field tests
• Manual vs. manual video
• Manual vs. manual video vs. passive infrared
• Used correlation as proxy for accuracy

9
Field Test 1 San Francisco

• 10 intersections
• 4-5 hours each
• One field vid observer
• Platoons of pedestrians crossing
• Multiple attributes

10
Field test 2 City of Berkeley

• Three sites for four hours each
• One field observer (Sally) and one video observer
• One automated passive infrared pedestrian counter
• Counted sidewalk line crossings with clicker
• Counts only -- no attributes

11
Field test 2
12
The results

• Field test 1
• Field observer systematically undercounted
• No relationship to volume
• Undercount with clickers, but sample small
• Field test 2
• Passive infrared systematically undercounted
• Field observer / video produced same count
• No relationship to volume

13
Issues

• Why different accuracy results for field observer
  vs. video?
• Different counting task
• Observer motivation
• Why is volume seemingly unrelated to accuracy?
• For manual observer motivation
• For automated proportion of adjacent peds

14
Conclusions
15
Further research needs

• Observer motivation accuracy
• Complexity of counting task accuracy
• Infrared, ped flow characteristics accuracy