Mining for meaning in driver’s behavior: A tool for situated hypothesis
generation and verification
E.R. Boer1, C.A. Joyce2, D. Forster2, M. Chokshi2, T. Mogilner2, E.
Garvey2 and J. Hollan2
1LUEBEC, San Diego, CA, USA
2Distributed Cognition and HCI Laboratory, Department of Cognitive
Science, University of California, San Diego, La Jolla, CA, USA
Driving is one of the most complex tasks we all share. Its complexity
is not attributed to the task itself but to the context within which the
task is performed. To combat or confront this complexity, researchers,
in the interest of time, either highly constrain the conditions of study
(human factors) or delve deep into the myriad of factors that influence
behavior for a limited number of drivers or situations (ethnography).
The ethnographic approach yields a situated understanding of how individuals
or teams produce meaningful behavior in a natural context. It is a lengthy
process due to the need to examine the data (i.e. video, time series,
notes, annotation, verbal reports, audio recordings, etc) from many perspectives
and mold these views into one coherent object which embodies the meaning
of the studied behaviors. In this paper we present a tool (‘CoBeX’)
developed to aid ethnographers by integrating visualizations of heterogeneous
data sources in a time or context synchronized fashion. The tool’s
main power is in visualizing and cross-referencing data occurring at the
same time point across many integrated and aggregated data representations,
thus offering a quick means to detect and understand apparent patterns
and outliers.
Much of science involves understanding the apparent consistency and variability
in an observed process through contextual exploration. CoBeX was designed
to support explorations into how context affects driver behavior and whether
this differs across drivers. Through iterative data augmentation and outlier
exploration of multidimensional data captured on a highly instrumented
vehicle, we located patterns that characterize driver behavior in lane
changing and car following. These were used to construct a complete model
that embodies the interactions between elements that produce the observed
behavior. By offering a means to accelerate the ethnographic process,
naturalistic driving tasks become a more feasible research option for
increasing our understanding of driver behavior and improving the design
of driver support systems.
Paper presented
at Measuring Behavior 2005
, 5th International Conference on Methods and Techniques
in Behavioral Research, 30 August - 2 September 2005, Wageningen, The
Netherlands.
© 2005 Noldus
Information Technology bv
|