Search Results

In 2010, 32,855 fatalities and over 2.2 million injuries occurred in automobile crashes, not to mention the immense economic impact on our society. Two of the four most frequent types of crashes are rear-end and lane departure crashes. In 2011, rear-end crashes accounted for approximately 28% of all crashes while lane departure crashes accounted for approximately 9%. This paper documents a study on the NADS-1 driving simulator to support the development of driver behavior modeling. Good models of driver behavior will support the development of algorithms that can detect normal and abnormal behavior, as well as warning systems that can issue useful alerts to the driver. Several scenario events were designed to fill gaps in previous crash research. For example, previous studies at NADS focused on crash events in which the driver was severely distracted immediately before the event. The events in this study included a sample of undistracted drivers.

Distracted driving remains a serious risk to motorists in the US and worldwide. Over 3,000 people were killed in 2013 in the US because of distracted driving; and over 420,000 people were injured. A system that can accurately detect distracted driving would potentially be able to alert drivers, bringing their attention back to the primary driving task and potentially saving lives. This paper documents an effort to develop an algorithm that can detect visual distraction using vehicle-based sensor signals such as steering wheel inputs and lane position. Additionally, the vehicle-based algorithm is compared with a version that includes driving-based signals in the form of head tracking data. The algorithms were developed using machine learning techniques and combine a Random Forest model for instantaneous detection with a Hidden Markov model for time series predictions.