In the right hands, automated systems like computer vision and artificial intelligence can act as tools for humans to make better decisions. For transportation planners and advocates, they can lead to better use of resources and better distribution of space for all road users.
At Transportation Techies’ Playing with Traffic Meetup, locally- and internationally-focused coders shared how they have combined machine learning with human analysis to build safer streets around the world.
A bot that aggregates unpaid driving citations
Daniel Schep and Mark Sussman shared their Twitter bot, @howsmydrivingDC. When someone tweets a license plate number at the bot, typically in response to the car’s driver behaving dangerously, the bot accesses Washington, DC’s Department of Motor Vehicles’ (DMV’s) database on unpaid tickets for parking and automated enforcement – like speed cameras – infractions. The bot then replies to the tweet with a spreadsheet of the vehicle’s entire tab of unpaid tickets.
The #BikeDC community has interacted enthusiastically with the bot due to the high rate of cars blocking bike lanes in the District, and this has led to a wealth of data for Schep and Sussman to explore. From 459 tweets, the bot has uncovered over $300,000 in unpaid fines from 2,069 citations.
That goes to show how people tend to be repeatedly dangerous, and many of the license plates the bot scanned had a growing list of citations, with many owing over $5000 each. If there is no other lesson from this, according to Schep, it’s at least that if you feel like the person blowing through a red light is a jerk, they actually likely are.
Mark Sussman presenting. Photo by Michael Schade.
Beyond being a cathartic project, as Schep described it, he and Sussman want to grow this into a robust partnership with the DMV. With funding they hope to win from the GigabitDCx competition, they hope to develop an automated scanner of photos of license plates that then communicate with enforcement officials so they know where to find the repeat offenders that make roads unsafe.
While behavior is a significant factor in dangerous driving, the built environment can enable or discourage some of the worst infractions. As a result, @howsmydrivingDC could pair well with a new database from the District Department of Transportation (DDOT), roadway cross-sections, which James Graham shared. According to Graham, DDOT came to understand that traditional data practices did not provide robust enough information to perform the data analysis that DDOT and advocates hoped to accomplish.
The new database hopes to improve the agency’s understanding of roadway conflict by providing information on all roadway characteristics in the District to include the number of lanes, their widths, their use, and even the surface material. Adding this information to crash data, as Graham used in an example, would not only tell DDOT where crashes are occurring but could point to why they are occurring, especially in high-risk spaces.
DDOT can now follow crash data to the worst areas and then add road characteristics to understand the underlying factors that may be contributing to higher risk in a particular space.
Together, tools like @howsmydrivingDC and DDOT’s cross-sectional data could combine to help the District better understand where to target safety enforcement efforts in real time, and over the long term to target infrastructure improvements. They would provide important levels of understanding that can guide decisions in pinpointing how limited resources would be best used.
Heading off the beaten path
Some Techies from much larger organizations shared a number of tools they use in international development to manage roads. Though they may have been built around development and disaster recovery, there is potential for cities like DC to adapt to local efforts for improving resource allocation for infrastructure repair.
Dan Joseph from the American Red Cross showed how his organization uses vehicle-mounted cameras to map street conditions for disaster preparedness, response, and recovery. Using machine learning with human verification, the photos they take – about once every second – help the Red Cross build a 3D map of their focus area to understand where to focus efforts before a disaster, particularly for informal settlements where many people can go missing.
In addition, using these cameras while responding to disasters helps responders evaluate where to focus their efforts and how they can do so based on the extent of damages to the roads they need to use.
The 3D maps, with before and after comparisons, can be vital for disaster response anywhere, but could have some applications in places like DC for more mundane situations, like studying how to improve bus routes. Perhaps it would work to record conditions that buses travel through, and combined with DDOT’s new database could better evaluate movement through trouble spots to redesign the streets in transit’s favor.
Holly Krambeck presenting. Photo by Michael Schade.
Holly Krambeck of the World Bank showed how the organization built a network of mappers to create a national inventory of streets and their conditions across Laos. The effort stemmed from a need to understand which roads were paved since people who live along dirt streets are cut off from the supply chain during floods.
Using photography of roads similar to the Red Cross’ approach, Krambeck’s team then had community mappers tag the images with a range of details, including paving status and any potholes they saw. Once these data were verified, the teams edited the street information in Open Street Map, which Laos officials can use to identify high-need areas, and which advocates can use to draw attention to those same spaces.
Working with the machines
It would be nearly impossible for humans to effectively process all of the data these tools generate without automated assistance. Yet collecting such a large, granular set is what could make a difference in developing efficient infrastructure. Machine learning can do this – it’s a tool that leaves decisions in human hands – it just categorizes and identifies the things we’re already looking for in a much faster and more expansive manner.
As communities continue their push to prioritize road safety and mobility over car movement, they will need the tools to understand what changes would have the most impact. From direct participation and direction on dangerous road users to deeper understanding of existing infrastructure, planners and advocates can leverage tools like those presented at Transportation Techies to make a difference.
The post Can machine learning improve road safety and traffic? Techies try it out appeared first on Mobility Lab.