With nearly a billion automobiles on the road today, and a doubling projected over the next decade, transportation provides indispensable functions in support of meeting the basic needs of society for mobility and accessibility, and directly affects the quality of life. However, with the continued increase in travel demand and non-sustainable development patterns, transportation systems have begun to show signs of serious strain, such as congestion, traffic accident, excessive energy consumption and increased emission level. To mitigate these problems, aside from developing more efficient and intelligent automobiles, making the acquisition and spread of road/traffic information more precisely, efficiently, and timely is of equal, if not more, importance.
Traditionally, road and traffic monitoring are conducted through either stationary sensors placed on or by the side of the road, such as inductive loop detectors and closed-circuit cameras, or specialized probe vehicles equipped with dedicated sensors, such as ground penetrating radar (GPR) and 3D laser scanner. Unfortunately, the prohibitively high deployment cost of such devices makes it impossible to achieve largescale deployment, leading to limited road coverage and delayed information update.
The recent proliferation of increasingly capable and affordable mobile devices (e.g., smartphones) packed with a plethora of on-board sensors (e.g., GPS, accelerometer, compass, camera, etc.) makes this problem tractable. Now we can outsource the road sensing tasks to a large crowd of individual vehicles with human-carried mobile devices. Thus far, a wide spectrum of such crowdsourced road sensing systems have been developed, monitoring various road events, such as congestions, accidents, potholes, bumps, and road signs. However, human-carried mobile devices, compared to dedicated road monitoring sensors, are not reliable, due to various reasons such as poor sensor quality, lack of sensor calibration, background noise, and incomplete views of observations. They may provide inaccurate or even false information that could mislead people’s decisions, and eventually result in invaluable loss. To address this sensor reliability problem, we can integrate the information from multiple vehicles that observe the same road events, as this will likely cancel out the errors of individual vehicles and improve the event detection/estimation accuracy.
A straightforward way to integrate the sensory data from multiple vehicles is to simply take voting/averaging on them. This approach, however, treats all the vehicles equally, and fails to capture the variety in the Quality of Information (QoI) of different vehicles. Intuitively, if we can identify and put more weights on the vehicles with high quality, the integration accuracy can be significantly improved. To achieve this, a major challenge has to be addressed. That is, the QoIs of individual vehicles are usually unknown a priori. We have to infer QoIs from data after the road events are reported. To tackle this challenge, in this project, we propose CrowdRoad, a quality-aware crowdsourced road sensing system.