Cover Image

Not for Sale

View/Hide Left Panel
Click for next page ( 46

The National Academies of Sciences, Engineering, and Medicine
500 Fifth St. N.W. | Washington, D.C. 20001

Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement

Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 45
45 contributing factors, such as weather, traffic flow, traffic den- selection of threshold values is always a compromise between sity, traffic control, road profile, surface condition, and light- exhaustiveness and false events. Lower trigger values will ing condition, are identified. If available from the video clip, capture the maximum number of potential events, but the the status of other involved vehicles can also be examined and trade-off is a higher chance of false-positive events, nonconflict coded. A similar procedure is applied to reduce baseline epochs. events, and less severe conflicts. Similarly, a higher trigger value The baseline epochs were used as a benchmark in which driver will result in a higher percentage of valid events but will gener- behavior can be compared with that in a safety-related event. ate some omissions. A careful examination of threshold values According to the design of the researchers, a certain number by experts is highly recommended before data reduction starts. of baseline epochs of selected length are decided. For exam- For studies such as Project 2 and Project 5, which were ple, the baseline epochs in Project 7 were 60 s long with a total aimed at testing onboard alarm systems, data reduction is rel- of 1,072 epochs, whereas in Project 8 the baseline epoch was atively more straightforward. When an alert was triggered or 30 s long and 456 were selected. the driver comment button was pushed, the main system To ensure high-quality data reduction, a quality control logged the triggered summary file, and the video system was procedure needs to be established. On the one hand, differ- notified and captured a retrospective clip of video data with ences between data reductionists should be minimized. On transition counts, histograms, errors, and other trip summary the other hand, data reductionists should ensure that their information recorded to a trip-summary log. Because data judgments are consistent over time. Accordingly, inter-rater were already organized by trips and alarms during collection, and intra-rater reliability are tested regularly. Before a data the scanning step to identify potential events from numerical reduction effort starts, several test events are selected and data is unnecessary. Data reductionists can start data reduc- coded by expert analysts (e.g., the principal investigator [PI] tion from viewing epochs of video data of events and then of the project or a data reduction director who has extensive coding variables to the data dictionary. Data associated with experience with data reduction). Next, each reductionist is the reduced data include demographic information (driver's asked to code the same events. Their coding results are com- age-group), lighting condition, road type, and traffic density. pared with those of the experts, and discrepancies are noted Additionally, numerous secondary behaviors (behaviors and discussed. The validation results help to determine (1) if besides driving) are coded and include such actions as con- reductionists are correctly coding the events; (2) if certain versing, adjusting the radio, speaking on a cell phone, and reductionists have systematic bias on certain questions; and brushing one's hair. The same modeling method for estimat- (3) if the data dictionary is rationally composed and easily ing travel time reliability and judging if driver behavior is cor- understood. Associated training, supervision, and modifica- rectable can be applied. Figure 5.1 summarizes a typical data tion of the data dictionary are applied if needed. For the intra- reduction process (2). rater reliability test, one example event is selected for each category representing crash, near crash, proximity, and crash- relevant events. Data reductionists are required to code these References events to the dictionary at the beginning of the process. They 1. Hanowski, R. J., M. Blanco, A. Nakata, J. S. Hickman, W. A. Schaudt, code the same event again after a period of time (varying from M. C. Fumero, R. L. Olson, J. Jermeland, M. Greening, G. T. Hol- a week, a month, or a year, depending on the full length of the brook, R. R. Knipling, and P. Madison. The Drowsy Driver Warning data reduction effort). The results are compared with the System Field Operational Test, Data Collection: Final Report. Report original reduction results. If there are differences, reduc- DOT HS 811 035. NHTSA and Virginia Tech Transportation Insti- tute, Blacksburg, Va., 2005. tionists are notified to review their work and make necessary 2. Lerner, N., J. Jenness, J. Singer, S. G. Klauer, S. Lee, M. Donath, M. adjustments. Manser, and N. Ward. An Exploration of Vehicle-Based Monitoring of In summary, three-step data reduction is effective in pro- Novice Teen Drivers: Draft Report. Virginia Tech Transportation cessing large-scale video data and other numeric data. The Institute, Blacksburg, Va., 2008.