Powering AI Through Crowdsourcing #2: Preventing Traffic Deaths in Washington, D.C.


Powering AI Through Crowdsourcing #2: Preventing Traffic Deaths in Washington, D.C.

By Adam Karides

Road traffic is the 9th leading cause of death every year. It is responsible for taking 1 million lives around the world each year, 30,000 of which occur in the United States. To combat this growing trend, Vision Zero has emerged as a collection of local, state, and national governments committed to eliminating all traffic-related deaths by 2024. It was established by the Swedish government in 1997, which has maintained one of the lowest traffic-related death rates in the world. However, other governments and communities are late to the game, including Washington, D.C., which did not join until December 2015 . But the nation’s capital is working to catch up by converging crowdsourcing, artificial intelligence, and open data.

The District’s Department of Transportation (DDOT), led by Mayor Muriel Bowser, announced its heightened commitment to Vision Zero by joining the initiative’s Video Analytics program, which joins forces with Microsoft to better leverage the data collected by cameras the city has already installed to record traffic patterns to prevent deaths. In particular, it aims to recognize and predict potential hazards by registering and feeding these recordings into artificial intelligence software, but DDOT does not have the manpower to sift through all of these cameras’ recordings and label all of the moving objects. For these machines to effectively analyze and identify these potentially harmful traffic patterns, the program is crowdsourcing individuals to train them.

Individuals contribute to this program by watching video clips recorded by traffic cameras, tagging the objects captured in the clips, and tracing their movements in the frame. By labeling the pedestrians, bicycles, and vehicles throughout a given clip, the AI software linked to these cameras will be able to identify these moving parts in real time. So how does this program save lives? By identifying and tracking the objects at a particular intersection, the cameras will be able to recognize patterns of movement that indicate a higher likelihood of a traffic accident. Traffic lights then adjust their signals accordingly to redirect the flow of traffic in the least dangerous way possible.

Currently, transit officials attempt to reduce traffic-related deaths by analyzing previously reported traffic data and tweaking the traffic control programs accordingly. However, these actions fall short of being able to predict and prevent these threats in real time. Traffic systems must be able to identify potential hazards as they unfold, and by crowdsourcing these machine learning protocols, the individuals with the most knowledge of these particular intersections are able to optimize them. In the words of Mayor Bowser, “Residents know traffic issues in their neighborhoods better than anyone, and now we will be able to leverage their knowledge with our existing camera infrastructure in order to prevent crashes and injuries before they occur.” Our traffic systems must be preventive, not reactionary in order to save lives and eventually achieve Vision Zero’s objectives, and crowdsourcing is the engine that will save these lives.

The Video Analytics Towards Zero Vision project is another example of an overarching understanding of why humans are necessary to render functional and effective AI systems that benefit our society. But a handful of individuals cannot fundamentally drive sustainable machine learning; human input must operate at large scale, which is best attained through crowdsourcing. Furthermore, projects such as this one are not made possible without an emphasis on open data, which underscores these crowdsourcing efforts by empowering individuals with the necessary tools and resources to train machines, and more importantly save lives.

There is a prime opportunity to embrace artificial intelligence technologies to enhance our traffic control systems, but their performance is augmented by human input that maintains and sharpens this software to eventually prevent deaths. These preventive measures will be even more critical when self-driving cars hit the pavement, which are using their own crowdsourcing tactics to build out similar recognition features, to develop the necessary technological harmony between vehicles on the road and the systems that conduct their flow in order to fulfill Vision Zero. To participate in Video Analytics Towards Vision Zero campaign, please watch this instructional video, and start labeling!


Didn’t get a chance to read the first article of this series, or just want to stay up-to-date with later editions and other news and trends? Sign up for our monthly newsletter here!