![]() | Only 14 pages are availabe for public view |
Abstract It has been noticed over the past few years that driver distraction has been responsible for good percentage of road accidents. Moreover, cell phone usage has been reported to be a major cause for driver distraction-based road accidents. These observations led governments to ban cell phone usage among distracted drivers. The new regulations for banning cell phone usage while driving raised the need for automated systems to monitor drivers and report cell phone usage violations. Previous studies in this area show some systems for detecting driver distraction based on computer vision techniques and other systems based on vehicle behavior. Among the computer vision based systems, some are intended for alerting distracted driver and hence are mounted inside the vehicles. While others are roadside setup to report cell phone usage violations. We propose an integrated system to automatically detect, track, and report distracted drivers. Based on real time camera feed, our Convolutional Neural Networks (CNN) detect vehicles drivers and hence classify their attention. An automatic license plate recognition module records distracted drivers’ vehicles’ plate numbers and report them through a simple web-based server component. To train and test our CNNs, we collected and annotated various video footage of volunteering drivers from a camera mounted on a patrol car as well as a roadside gantry. We collected two ix datasets for the roadside and patrol setups. Then, we trained a CNN based classier to classify drivers’ behavior as whether they are driving safely or using their cell phones. Our system achieved 89% driver detections and 95% classication accuracy on a recorded test set. We also report the results of one full day run of the roadside setup which resulted in 100% recall of cell phone usage and 46% precision and discuss these results. |