FHWA EAR Project: Automated Feature Extraction Data Sets: Event Record and Seatbelt Usage Data

Project Description Oak Ridge National Laboratory (ORNL) is functioning as a technical advisor and independent system validator for the Federal Highway Administration for Broad Agency Agreement (BAA) call DTFH61-13-R-00011, “Exploratory Advanced Research Program”. In this document, we refer to BAA participants as “Performer Teams” (PT). One particular area of interest in the BAA where ORNL will assist is the Automated Feature Extraction (AFE) of naturalistic driving data, from available video streams and other sensor information. AFE refers to algorithms that process video data and automatically extract behavior characteristics of the driver and surroundings. There are several examples of features which could be extracted automatically, such as: the number of passengers in the vehicle with the driver; the traffic density of the surrounding vehicles; outside weather conditions; head pose; driver drowsiness; etc. Many of these features are derived from lower-level image analysis algorithms that detect motion, objects of interest, and other events in video. Due to the large size of the data set, automating the coding of the data through AFE can yield tremendous value to safety research and dramatically improve the utility of the SHRP-2 NDS data. The purpose of this analysis is not directly concerned with safety analysis or attempting to find solutions or study a particular safety or traffic problem. Instead, we are striving to develop automation methods, which, in turn, will simplify the tasks of finding data in the NDS to solve traffic problems. Data Specification Dataset/Excel spreadsheet with event detail table from InSight (September 2014 release), MP4 forward video files for said events, CSV files with time series data for said events