About the research
Wrong way driving crashes occur infrequently, accounting for almost 3 percent of all crashes, but they have a very high likelihood of resulting in fatal or serious-injury crashes. The causes associated with wrong-way crashes tend to make them spatially concentrated to particular stretches of roads, making it important to identify and monitor such high-risk locations.
This research will automate wrong-way detection using existing closed-circuit television (CCTV) cameras. In the past few years, there has been a significant revolution in the field of image processing. This has been due to the emergence of hardware called a graphics processing unit (GPU) cluster and a machine-learning technique called deep-learning. Deep learning has revolutionized the field of multi-object detection and classification. Hence, this is an ideal time to revisit the issue of using CCTV cameras as a traffic sensor.
Vision-based solutions are attractive as they also provide images for the operators to manually validate the alarms. The Iowa Department of Transportation (DOT) has deployed over 365 cameras on their freeway system, which can be accessed over the network. This study will develop software to estimate wrong-way likelihood using those existing cameras. If the automation technology proves viable, these connected cameras can be converted into a wrong-way detection system that could identify high-risk locations even before the crash history develops.