Using Webcams and Crowds to Study Active Transportation

(Originally published on 09/23/15)

By J. Aaron Hipp, Alicia Manteiga, Amanda Burgess, Abby Stylaniou, and Robert Pless

Abstract

Active transportation opportunities and infrastructure are an important component of a community’s design, livability, and health. Features of the built environment influence active transportation, but objective study of the effects of built environment improvements on active transportation is challenging. Outdoor temperature is known to be a common barrier to active transportation, yet there is limited information examining the interaction between temperature, built environment improvements, and active transportation. In this case study, 20,529 publicly available webcam images from two intersections in Washington, D.C., were used to examine the impact of an improved crosswalk on active transportation. A crowdsource, Amazon Mechanical Turk, analyzed image data. Temperature data was collected from the National Oceanic and Atmospheric Administration. Summary analyses demonstrated slight, bi-directional differences in the number of images with pedestrians and bicycles captured before and after the enhancement of the crosswalks. Chi-square analyses revealed these changes were not significant. In general, pedestrian presence increased in images captured during moderate temperatures compared to images captured during hot or cold temperatures. Chi-square analyses of one intersection indicated the crosswalk improvement may have encouraged walking in more uncomfortable outdoor temperatures. The methods employed provide an objective, cost-effective alternative to traditional means of examining the effects of built environment changes on active transportation. The use of webcams to collect active transportation data has applications for community policymakers, planners, and health professionals. Future research should work to validate this method in a variety of settings as well as across different built environment and community policy initiatives.


Introduction

Active transportation, bicycling and walking between destinations, is associated with reduced rates of chronic disease and the promotion of healthier lifestyles in comparison to vehicle trips.1-5 Some characteristics of the built environment influence the adoption of behavioral changes, encouraging individuals to choose walking or biking over personal motor vehicle use.1-3, 5-7 Characteristics such as perceived safety, proximity to destinations, the presence of foliage and green spaces, and traffic control features like stop signs and speed bumps are mediating factors in this decision making process.1-3, 5-9

The improvement of infrastructure along transportation routes can impact the decision to walk or ride a bicycle in a community.5, 6, 8, 9 While these improvements are often associated with an increase in active transportation, there is little agreement on the specific elements that lead to this increase.9

Pedestrian safety has been at the forefront of a body of research evaluating built environment characteristics that aid or hinder the decision to walk.10-12 Most crosswalk enhancement studies focus on reducing pedestrian and vehicular collisions.10-12 Basic marked crosswalks are more effective than unmarked crosswalks at increasing pedestrian safety. Evidence indicates stand-alone crosswalks, independent of other interventions like speed limit reductions or speed bumps, reduced the number of intersection collisions across 30 cities in the United States .10 Beyond pedestrian safety, few studies have analyzed the effects of adding or enhancing crosswalks on pedestrian activity.

Permanent built environment features such as bicycle boulevards and improved bike lanes are associated with an increase in biking.7, 13-15 The relationship between factors influencing the use of biking spaces is complex, making it difficult to measure the effects of this infrastructure change on active transportation.7, 16

The decision to choose to walk or bike instead of drive is associated with weather as well as with features of the built environment.4, 6, 23, 24 Pedestrian activity decreases overall during snow, ice, and cold temperatures in winter seasons, but individuals are more likely to choose walking over biking when the temperature is cooler.4, 23-25  There is a lack of literature on the role temperature plays in the utilization of new built environment features in a community, which is likely due to the complex nature of the data needed to study such outcomes.

Understanding trends in pedestrian and bicyclist behaviors, especially in response to specific built environment interventions, allows key stakeholders to select policies for implementation that will have the greatest impact on their community’s specific active transportation needs. 4, 9, 17

The evaluation of built environment interventions requires active monitoring of these changes in outdoor spaces.4, 5, 9, 13, 17 The Archive of Many Outdoor Scenes (AMOS) is a database that has compiled images captured by publicly available webcams since 2006 (e.g., traffic cameras).14, 18 Images gathered from the AMOS database (over 735 million) can be analyzed to study changes in the built environment as well as changes in active transportation. The use of webcams to study active transportation can provide researchers and community stakeholders access to outcome data regarding the impact of built environment enhancements on active transportation.15, 19

The manual annotation of a large number of webcam images has the potential to be both time consuming and costly, but emerging technologies such as crowdsourcing can alleviate these constraints.20-22  Amazon’s Mechanical Turk (MTurk) is a crowdsourcing platform that allows anyone to design a human intelligence task (HIT) such as counting people in webcam images, and then post this HIT online for individuals over the age of 18 to complete via the internet.20, 22  Posting a HIT to the ‘crowd’ of human workers online allows researchers to obtain quality data quickly and inexpensively.20-22

The primary objective of this case study was to determine whether or not trends in active transportation are influenced by a change in the built environment. Additionally, the study aimed to explore interactions between built environment enhancements, temperature, and active transportation.  Finally, we sought to present a novel method for objectively and inexpensively measuring the effects of built environment changes on active transportation, thus building the bridge between health and community planning projects.


Methods

Active transportation data was from two webcams captured by the Archive of Many Outdoor Scenes (AMOS) dataset. Webcams used in this study are located at the intersections of Piney Branch Road NW and Eastern Avenue NW in Washington, D.C., 20012 (residential area, Figure 1), and Connecticut Avenue NW and Florida Avenue NW in Washington, D.C., 20009  (commercial area; Figure 1). The two intersection webcams were selected because they have a clear view of pedestrians, bicyclists, and vehicles, and because both captured the enhancement of crosswalks on November 20, 2007. Intersections were classified for general land use using Google Street View.26, 27

Using the AMOS dataset, images were collected every 30 minutes over an average of twelve hours per day, or approximately 24 images per day. In total, 20,529 webcam images were captured between May 7 and November 19, 2007, and between May 7 and November 19, 2008.

To examine the effects of temperature and precipitation on the use of crosswalks, images were matched with an hourly average temperature and precipitation status. Due to limited data availability, researchers only combined 8,067 images with hourly temperature data, and elected to ask MTurk workers about precipitation. Temperature and precipitation data was collected from the National Oceanic and Atmospheric Administration: National Centers for Environmental Information (http://www.ncep.noaa.gov/).

The Amazon Mechanical Turk (MTurk) website was used as the crowdsourcing platform to annotate the number of pedestrians and bicyclists in each captured image. Each image was annotated by four unique MTurk workers.15 MTurk workers were paid US $0.02 per image between September and December 2013. Some of the prompts MTurk workers responded to included:

  1. Please outline each bicycle or person riding a bicycle in the scene
  2. Please outline each pedestrian in the scene
  3. What is the weather in this image? Sunny, Cloudy, Rainy, Snowy

Each image was annotated at least four unique times (N=82,116), for a total cost of $1,642.32.

Counts per transportation mode were downloaded into SPSSv.22 (IBM) for analysis in March 2015.

Summaries of weekday, weekend, and overall presence of pedestrians and bicyclists were calculated at both intersections, and have been reported elsewhere.19 Chi-square analyses were performed to study differences in pedestrian and bicyclist presence in images before and after crosswalk enhancement.

Average and standard deviations of collected temperature data were calculated. Scatterplots were created to visually portray the relative frequency of pedestrians per intersection at various temperatures. Temperatures were divided into “normal” (within one standard deviation of the May-November mean temperature) and “non-normal” (outside one standard deviation of the mean temperature) categories. Summaries of the number of images with pedestrians and bicyclists, prior to and following crosswalk enhancement, at normal and non- normal temperatures were calculated. Chi-square analyses were performed to determine if there were differences in bicyclist and pedestrian presence when temperatures were normal versus non-normal.

Descriptive summaries of the number of images MTurk workers identified as rainy were compiled to determine the degree of agreement between crowdsource workers.


Results

Descriptive Statistics

At the residential intersection, 4,959 images were captured prior to and 5,007 images were captured following the crosswalk enhancement. Pedestrians were present in 298 images (6.01 percent of all images) captured prior to the change and in 337 images (6.73 percent) captured following the change. Bicycles were present in 79 images (1.59 percent) captured before the change, and in 86 images (1.72 percent) captured after the change.

At the commercial intersection, 5,246 images were captured prior to the crosswalk enhancement and 5,317 images were captured following the crosswalk enhancement. Pedestrians were annotated in 3,658 images (69.73 percent) captured prior to the change, and in 3,615 images (67.99 percent) captured following the change. Bicycles were annotated in 581 images (11.08 percent) captured before the change, and in 565 images (10.63 percent) captured after the change (Table 1).

3,883 images at the residential intersection and 4,184 images at the commercial intersection were matched with temperature data. Temperatures ranged from 27 to 101 degrees Fahrenheit, with an average of 74 degrees, and a standard deviation of 12 degrees. In general, there were more pedestrians per image when temperatures were within one standard deviation of the mean (between 62-86 degrees) (Figure 2). Bicycle annotation patterns were not related to temperature at both intersections (Table 2).

Temperatures were divided into two categories: non-normal (less than 62 degrees or greater than 86 degrees) and normal (between 62 and 86 degrees, inclusive). At the residential intersection, before the enhancement of the crosswalk and when temperatures were non-normal, 32 images (6.61 percent) included pedestrians. After the enhancement when temperatures were non-normal, 55 images (8.44 percent) included pedestrians. There were 10 images (2.07 percent) and 13 images (1.99 percent) with bicycles before and after the change, respectively, when temperatures were non-normal.

At the commercial intersection prior to the crosswalk enhancement and when temperatures were non-normal, 373 images (67.70 percent) included pedestrians. This changed to 477 images (66.52 percent) after the change. Bicyclist annotation in images captured during periods of non-normal temperatures changed from 67 images (12.6 percent) to 78 images (11.61 percent) after the crosswalk enhancement (Table 3).

 

 

Researchers attempted to assess precipitation by asking MTurk workers the following question: What is the weather in this image: Sunny, Cloudy, Rainy, Snowy. These four options were collapsed into two categories, no precipitation or precipitation. Across only 55 percent of all images did all four MTurk workers agree on the image having precipitation or not. Therefore, researchers determined the wording of the question was not reliable, and elected to eliminate precipitation from analyses in this study.

Chi-Square Analyses

Chi-square tests of independence were performed to examine the relationship between presence of pedestrians and bicycles in images before and after the enhancement of crosswalks (Table 1). The overall (weekday and weekends combined) relationship between pedestrian presence and crosswalk enhancement at the residential intersection was not significant [X2(1, N= 9,966) = 2.17, p=.14]. The overall relationship between bicycle annotation and crosswalk enhancement was also not significant [X2(1, N= 9,966) = 0.24,p=0.63], though there was a significant decrease in bicycle annotation during weekends after the crosswalk enhancement [X2(1, N=2,735) = 4.04, p=0.04]. At the commercial intersection, the relationships between crosswalk enhancement and pedestrian presence, and crosswalk enhancement and bicycle presence were not significant overall [X2(1, N=10,563) = 3.73, p=0.05; X2(1, N= 10,563) = 0.55, p=0.46]. However, there was a significant decrease in pedestrian presence during weekdays after the crosswalk enhancement [X2(1,N=7,599) = 4.08, p=0.04].

Chi-square tests of independence were then performed to examine the relationship between presence of pedestrians and bicycles in images before and after the enhancement of crosswalks at both normal and non-normal temperatures (Table 3). At the residential intersection prior to the crosswalk enhancement, there was no relationship between pedestrian annotation and temperature [X2(1, N=1,985) = 0.81, p>0.05], or between pedestrian annotation and temperature after the improvement of the crosswalk [X2(1,N=1,988) = 0.02, p>0.05]. There was no relationship between bicycle annotation and temperature, both prior to and following the crosswalk enhancement [X2(1, N=1,985) = 1.43, p>0.05; X2(1, N=1,988) = 0.22, p>0.05].

At the commercial intersection prior to the crosswalk enhancement, there were significantly more images with a pedestrian present during normal temperatures than during non-normal temperatures [X2(1, N=2,064) = 4.06, p<0.05]. However, after the completion of the crosswalk enhancement, there was no significant relationship between pedestrian presence and temperature [X2(1, N=2,120) = 0.22, p>0.05]. There was no relationship between bicycle annotation and temperature, both prior to and following the crosswalk enhancement [X2(1, N=2,064) = 0.13, p>0.05; X2(1, N=2,120) = 0.45, p>0.05].

Discussion

The results of this study indicate that two webcams in Washington, D.C., were able to capture pedestrian and bicyclist activity before and after the enhancement of two crosswalks, and across a range of temperatures. Pedestrian and bicycle annotation was not significantly different before and after the crosswalk improvement at both locations. An improved crosswalk may signal to drivers that there are non-drivers present, including walkers and cyclists. Therefore, it is unclear why pedestrian and bicycle annotation did not increase after crosswalk improvement. Potential explanations include increased vehicular traffic due to other improvements, or unsafe crosswalks along the way to the improved crosswalks.28 Future research could include a more broad analysis of a network of crosswalks. Such research may help explain variations in pedestrian presence in crosswalks after improvements, as well as establish which types of improvements are associated with the greatest increase in pedestrian activity.

Webcam images reflected temperature-related differences in pedestrian activity. Fewer pedestrians were annotated in images captured when temperatures were cold or hot. At both intersections after the enhancement, more images contained pedestrians captured during non-normal temperatures compared to the year prior. Furthermore, at the commercial intersection, the relationship between pedestrian presence and non-ideal temperatures prior to the crosswalk enhancement was significant; following the improvement of the crosswalk, the relationship was not significant. This suggests that the crosswalk may have played a larger influence on pedestrian presence than ambient temperature. Or stated another way, the addition of the crosswalk diminished the change in pedestrians between ideal temperatures and non-ideal temperatures. This may be due to a sense of increased sense of pedestrian safety or speed in crossing when temperatures were less than ideal.

In this study, sufficient hourly precipitation data was not accessible. Researchers attempted to identify precipitation visually using MTurk workers. The collection of reliable precipitation annotation by MTurk workers in images was a challenge. Researchers should continue to develop and validate weather-related image questions for crowdsourcing tasks. The presence of pedestrian and bicycle activity during inclement weather are of interest to community stakeholders invested in safety and transportation.12

Limitations of the present analyses include the use of only two intersections. The images used for analyses only provide information on behaviors at two specific locations, restricting the external validity of the findings. This study was unable to determine whether or not pedestrians were changing their routes.

Despite these limitations, the ubiquity and unobtrusive nature of webcams presents an opportunity to understand the effects of a variety of built environment improvements, across time, and in a cost-effective manner. While the applications of this method are still being fully developed, there is great promise in its potential. Potential applications include understanding which populations are benefitting from built environment enhancements, as well as broader studies examining the synergistic effects of multiple built environment changes.

Proximity to a built environment intervention, such as a crosswalk addition, does not necessarily indicate an impact will be made on nearby community members.7, 13 Webcams could be used to examine the influence of built environment changes on specific population groups such as adolescents or older adults. These populations generally have different motivating factors for participation in active transportation, and may receive more benefits from tailored built environment features than the general population.1, 3, 5, 6, 8, 17, 23, 24, 29

Future webcam research should also include the simultaneous analysis of multiple (e.g., greater than two) webcam locations in order to establish the external validity of the method. Studies should not be restricted to crosswalk enhancement or bike lane addition,14 but could include speed bump additions, median enhancements, or other environmental improvements relevant to specific communities. There is also the opportunity to work with civic and community partners in identifying non-built pedestrian safety improvement efforts, such as neighborhood watch groups and speed limit reductions.

In conclusion, the use of webcams and crowdsourcing is a promising technique for evaluating the effects of built environment interventions and environmental factors such as temperature on active transportation. As the method continues to develop, it is crucial that researchers and practitioners across community health and planning fields collaborate to explore various environments, interventions, and healthy behaviors.


 

References

1.            Hino AA, Reis RS, Sarmiento OL, Parra DC, Brownson RC. Built environment and physical activity for transportation in adults from Curitiba, Brazil. J Urban Health. Jun 2014;91(3):446-462.

2.         Winters M, Brauer M, Setton EM, Teschke K. Built environment influences on healthy transportation choices: bicycling versus driving. J Urban Health. Dec 2010;87(6):969-993.

3.         Panter J, Griffin S, Dalton AM, Ogilvie D. Patterns and predictors of changes in active commuting over 12 months. Prev Med. Dec 2013;57(6):776-784.

4.         Yang Y, Diez Roux AV, Bingham CR. Variability and seasonality of active transportation in USA: evidence from the 2001 NHTS. Int J Behav Nutr Phys Act.2011;8:96.

5.         Frank LD, Sallis JF, Conway TL, Chapman JE, Saelens BE, Bachman W. Many Pathways from Land Use to Health: Associations between Neighborhood Walkability and Active Transportation, Body Mass Index, and Air Quality. Journal of the American Planning Association. 2006;72(1):75-87.

6.         Mitra R, Faulkner G. There’s No Such Thing as Bad Weather, Just the Wrong Clothing: Climate, Weather and Active School Transportation in Toronto, Canada. Canadian Journal of Public Health. 2012;103(Supplement 3):S35-S41.

7.         Dill J, McNeil N, Broach J, Ma L. Bicycle boulevards and changes in physical activity and active transportation: findings from a natural experiment. Prev Med. Dec 2014;69 Suppl 1:S74-78.

8.         Grow HM, Saelens BE, Kerr J, Durant NH, Norman GJ, Sallis JF. Where are youth active? Roles of proximity, active transport, and built environment. Med Sci Sports Exerc.Dec 2008;40(12):2071-2079.

9.         Saelens BE, Handy SL. Built environment correlates of walking: a review. Med Sci Sports Exerc. Jul 2008;40(7 Suppl):S550-566.

10.       Zeeger CV, Esse CT, Stewart JR, Huang H, Lagerwey P. Safety Analysis of Marked Versus Unmarked Crosswalks in 30 Cities. ITE Journal. 2004:34-41.

11.       Huang H. An Evaluation of Flashing Crosswalks in Gainesville and Lakeland. In: Transportation FDo, ed; 2000.

12.       Cafiso S, Garcia Garcia A, Cavarra R, Romero Rojas MA. Crosswalk Safety evaluation using a Pedestrian Risk Index as Traffic Conflict Measure. 3rd International Conference on Road Safety and Simulation. Indianapolis, IN, USA; 2011.

13.       Hipp JA, Eyler AA, Kuhlberg JA. Target population involvement in urban ciclovias: a preliminary evaluation of St. Louis open streets. J Urban Health. Dec 2013;90(6):1010-1015.

14.       Hipp JA, Adlakha D, Eyler AA, Chang B, Pless R. Emerging technologies: webcams and crowd-sourcing to identify active transportation. Am J Prev Med. Jan 2013;44(1):96-97.

15.       Hipp JA, Adlakha D, Gernes R, Kargol A, Pless R. Do You See What I See: Crowdsource Annotation of Captured Scenes. SenseCam 2013. San Diego, CA, USA; 2013.

16.       Fernhall B, Borghi-Silva A, Babu A. The Future of Physical Activity Research: Funding, Opportunities and Challenges. Progress in Cardiovascular Disease.2015;57(4):299-305.

17.       Marzoughi R. Teen travel in the Greater Toronto Area: A descriptive analysis of trends from 1986 to 2006 and the policy implications. Transport Policy. 2011;18(4):623-630.

18.       Pless R, Jacobs N. AMOS: The Archive of Many Outdoor Scenes. Available at: http://amos.cse.wustl.edu/.

19.       Manteiga A, Hipp A. Learning from Outdoor Webcams: Capturing Active Commuting Behavior Across Environments. Paper presented at: Active Living Research, 2015; San Diego, California.

20.       Buhrmester M, Kwang T, Gosling SD. Amazon’s Mechanical Turk: A New Source of Inexpensive, Yet High-Quality, Data? Perspectives on Psychological Science. 2011;6(1):3-5.

21.       Kim AE, Lieberman AJ, Dench D. Crowdsourcing data collection of the retail tobacco environment: case study comparing data from crowdsourced workers to trained data collectors. Tob Control. Mar 2015;24(e1):e6-9.

22.       Amazon Mechanical Turk: Artificial Artifical Intelligence. Available at: https://www.mturk.com/mturk/welcome.

23.       Li Y, Hsu JA, Fernie G. Aging and the use of pedestrian facilities in winter-the need for improved design and better technology. J Urban Health. Aug 2013;90(4):602-617.

24.       Saneinejad S, Roorda MJ, Kennedy C. Modelling the impact of weather conditions on active transportation travel behaviour. Transportation Research Part D: Transport and Environment. 2012;17(2):129-137.

25.       Academies TRBotN. National Cooperative Highway Research Program: Guidebook on Pedestrian and Bicycle Volume Data Collection. 2014;797.

26.       Rundle A, Teitler J, Bader M, Lovasi L, Richards C, Lovasi G. Using Google Street View to Implement Community Audit Tools: The Pedestrian Environment Data Scan. Presentation at: Active Living Research, 2010; San Diego, California.

27.       Wilson JK, Kelly C. Navigating Google Street View: A Guide to Conducting Audits of the Built Environment Using Google Street View. Paper presented at: Active Living Research, 2011; San Diego, California.

28.       Fitzpatrick K, Turner S, Brewer M, et al. Improving Pedestrian Safety at Unsignalized Crossings:TCRP Report 112/ NCHRP Report 562. Washington, D.C. 2006.

29.       Graham DJ, Hipp JA. Emerging technologies to promote and evaluate physical activity: cutting-edge research and future directions. Frontiers in Public Health. 2014;2.

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s