(468a) Changing Idling Behavior through Dynamic Air Quality and Idle Detection Messaging

Concentrated vehicle engine idling can cause microenvironments of poor air quality, and areas with high idling, such as schools or hospitals, are frequented by individuals at increased risk for negative impacts from poor air quality. Anti-idling signage and education have been used to influence driver idling behavior. However, these strategies have had limited success. In this study, we try a different approach – dynamic feedback about air quality and idling status. A similar approach, dynamic feedback on driver speed, has become commonplace and has effectively reduced speeding and traffic accidents in the vicinity of the speed display. Our study aims to evaluate how dynamic messaging about idling behavior and air quality affects driver idling choices. Our study entailed developing a system that integrates networked low-cost air quality sensor measurements and idling vehicle detection based on video and audio inputs to provide dynamic feedback to drivers. The dynamic feedback system comprises seven air quality sensor nodes, one video camera, six microphones, a central server, and two outdoor displays to provide feedback to the drivers regarding idling status and local air quality. The low-cost air quality sensor nodes measured carbon dioxide (CO2), carbon monoxide (CO), nitric oxide (NO), nitrogen dioxide (NO2), total volatile organic compounds (TVOC), and particulate matter (PM2.5). The video camera and the six microphones provide information about vehicle motion and engine noise. The team developed a machine-learning technique to identify idling vehicles from the audio and video inputs. The server then integrated the air-quality sensor measurements and idling status (from the audio and video inputs) to provide real-time feedback about vehicle idling and air quality in messages sent to the outdoor displays. We deployed our dynamic feedback system at a hospital's drop-off zone, where numerous vehicles frequently idle. This deployment lasted 15 days. Five days showed a static control message on the outdoor displays; five days showed a machine learning idle status-based dynamic message; and five days showed an air quality metric-based dynamic message. The machine learning algorithm performed well over the 15-day deployment, with an average idling label precision of 78.3% and an average non-idling precision of 96.6% compared to onsite notes. The dynamic message deployment days resulted a 40% decrease in CO2 concentration and a reduction in the number of idling vehicles. The reduction in combustion emissions suggests that our dynamic messaging system can positively influence driver idling behavior in an area that typically experiences concentrated vehicle idling.