Traffic signals are widespread in developed countries as they allow competing flows of traffic to safely cross busy intersections. Traffic signals, however, do take their toll. The stop-and-go movement pattern that they impose, increases fuel consumption by 17%, CO₂ emissions by 15%, causes congestion, and leads to increased driver frustration. Drivers can be assisted with a Green Light Optimal Speed Advisory (GLOSA) system. A GLOSA system advises drivers on the optimal speed they should maintain when heading towards a signalized intersection. Should drivers maintain this speed, then the traffic signal will be green when they reach the intersection, allowing the driver to cruise through.

—Koukoumidis et al.

Worldwide, only a of handful GLOSA systems have been deployed; their costly and often impractical deployment and maintenance, however, has hindered their widespread usage, the authors note. Recognizing the potential benefit of GLOSA, US and European transportation agencies have advocated for the integration of short range (DSRC) antennas into traffic signals as part of their long term vision for intelligent transportation systems. (Earlier post.) DSRC-enabled traffic signals would be able to broadcast their schedule to DSRC-enabled vehicles that are in range.

This approach also necessitates the cost of equipping traffic signals and vehicles with the necessary specialized computational and wireless communications infrastructure.

In this paper, we take an infrastructure-less approach to accessing traffic signal schedules. We propose, implement and evaluate SignalGuru, a software service that runs solely on mobile phones, predicting the traffic signal schedule without any direct communications from the traffic signals. Our mobile phones are mounted on the vehicle’s windshield, and use on-phone cameras to detect and determine the current status of traffic signals. Multiple phones in the vicinity use opportunistic ad-hoc communications to collaboratively learn the timing patterns of traffic signals and predict their schedule.

—Koukoumidis et al.

There are drawbacks to such an infrastructure-less approach, the authors note in their paper:

• Lack of loop detector information. SignalGuru works without access to such information, and must perform predictions solely based on the information that can be measured by available mobile phone sensors.

• Commodity cameras. The quality of smartphones’ cameras is significantly lower than that of high end specialized cameras used in computer vision and autonomous navigation.

• Limited processing power. Processing video frames to detect traffic signals and their status (red, yellow, green) takes significant computational resources. A traffic signal detection algorithm that runs on resource-constrained smartphones must be lightweight so that video frames can still be processed at high frequencies. The higher the processing frequency the more accurately SignalGuru can measure the duration of traffic signal phases and the time of their status transitions.

• Uncontrolled environment composition and false detections. Windshield-mounted smartphones capture the real world while moving. As a result, there is no control over the composition of the content captured by their video cameras. Results from one of the deployments suggested that the camera-based traffic signal detection algorithm can confuse various objects for traffic signals and falsely detect traffic signal colors. A misdetection rate of 4.5% can corrupt up to 100% of traffic signal predictions.

• Variable ambient light conditions. Still image and video capture are significantly affected by the amount of ambient light that depends on both the time of the day and the prevailing weather conditions.

• Need for collaboration. The traffic signal information that an individual mobile device senses is limited to its camera’s view angle. A device may not be able to see a far-away traffic signal, or may not be within view of the traffic signal for a long enough stretch of time. Collaboration is needed between vehicles in the vicinity (even those on intersecting roads) so that devices have enough information to be able to predict the schedule of traffic signals. Koukoumidis et al. focused on a completely infrastructure-less solution that relies solely upon opportunistic communication (ad-hoc 802.11g) among the windshield-mounted devices.

The phone cameras capture video frames, and detect the color of the traffic signal using SignalGuru’s detection module. Information from multiple frames is then used to filter away erroneous traffic signal transitions (transition filtering module). Nodes running the SignalGuru service broadcast and merge their traffic signal transitions with others in communications range (collaboration module). Finally, a merged transitions database is used to predict the future schedule of the traffic signals ahead (prediction module).

The prediction of the future schedule of traffic signals is based on information about past timestamped R→G transitions i.e., information about when the traffic signals transitioned from red to green in the current or previous cycles. The prediction is based on R→G transitions, as opposed to G→Y (green to yellow) transitions, because vehicle-mounted smartphones can witness and detect R→G transitions much more frequently; when the traffic signal is red, vehicles have to stop and wait till the signal turns green. As a result, it is quite likely that a vehicle will be at the intersection at the moment that the R→G transition happens and thus detect it.

In addition to testing SignalGuru in Cambridge, where traffic lights are on fixed schedules, the researchers also tested it in Singapore, where the duration of lights varies continuously according to fluctuations in traffic flow. In Cambridge, the system was able to predict when lights would change with an error of only two-thirds of a second. In suburban Singapore, the error increased to slightly more than a second, and at one particular light in densely populated central Singapore, it went up to more than two seconds.

The computing infrastructure that underlies the system could be adapted to a wide range of applications, the researchers suggest, such as capturing information about prices at different gas stations, about the locations and rates of progress of city buses, or about the availability of parking spaces in urban areas, all of which could be useful to commuters.

Our proposed schemes improve traffic signal detection, filter noisy traffic signal data, and predict traffic signal schedule. Our results, from two real world deployments in Cambridge (MA, USA) and Singapore, show that SignalGuru can effectively predict the schedule for not only pre-timed but also state of the art traffic-adaptive traffic signals. Furthermore, fuel efficiency measurements, on an actual city vehicle, highlight the significant fuel savings (20.3%) that our SignalGuru-based GLOSA application can offer. Given the importance of traffic signals, we hope that this work will motivate further research in their detection, prediction and related applications.

—Koukoumidis et al.

SignalGuru is a great example of how mobile phones can be used to offer new transportation services, and in particular services that had traditionally been thought to require vehicle-to-vehicle communication systems. There is a much more infrastructure-oriented approach where transmitters are built into traffic lights and receivers are built into cars, so there’s a much higher technology investment needed.

—Marco Gruteser, an associate professor of electrical and computer engineering in the Wireless information Network Laboratory at Rutgers University

Resources

• Emmanouil Koukoumidis, Li-Shiuan Peh, Margaret Martonosi (2011) SignalGuru: Leveraging Mobile Phones for Collaborative Traffic Signal Schedule Advisory (MobiSys’11, June 28–July 1, 2011, Bethesda, Maryland)