Exploring Naturalistic Driving Data for Distracted Driving Measures
-
2017-10-01
-
Details:
-
Creators:
-
Corporate Creators:
-
Corporate Contributors:
-
Subject/TRT Terms:
-
Publication/ Report Number:
-
Resource Type:
-
Geographical Coverage:
-
TRIS Online Accession Number:01650927
-
Edition:Final Report 02/16/15 - 02/15/17
-
Corporate Publisher:
-
Abstract:The SHRP 2 NDS project was the largest naturalistic driving study ever conducted. The data obtained from the study was released to the research community in 2014 through the project's InSight webpage. The objectives of this research were to (a) explore the content of this large dataset and perform statistical analysis to identify useful performance measures to detect distracted driving behavior, and (b) provide an outline for a crash index model that can be used to quantify the crash risk associated with distracted driving behavior. Time series data on driver GPS speed, lateral and longitudinal acceleration, throttle position, and yaw rate were extracted as five appropriate performance measures available from the NDS that could be used for the purpose of this research. Using this data, the objective was to detect whether a driver was engaged in one of three specific secondary tasks or no secondary task at all using the selected performance measures. The specific secondary tasks included talking or listening on a hand-held phone, texting/dialing on a hand-held phone, and driver interaction with an adjacent passenger. Multiple logistic regression was used to determine the odds of a driver being engaged in one of the secondary tasks given their corresponding driving performance data. The results indicated that while none of the models provided a statistically good fit of the data, the lateral acceleration measure seemed to be a useful indicator of drivers' engagement in talking/listening and texting/dialing on the cell phone. The analysis of distracted driving behavior for by age and gender showed slightly different results. The longitudinal acceleration variable appeared to perform better in predicting talking/listening and texting/dialing for drivers aged 70-89. The lateral acceleration measure, however, performed better in predicting the engagement of younger drivers (16-29) in the same secondary tasks. When considering the gender of drivers, the lateral acceleration performance variable proved to be more effective in predicting texting/dialing and talking/listening for both genders. Still, these results are inconclusive due to the undesirable Hosmer and Lemeshow Test p-values observed in all the models. Thus, the same analysis was performed using neural networks modeling which is recognized for its capability of nonlinear pattern recognition. The neural network analysis showed that the five performance measures can be used as surrogate measures of distracted driving. The developed neural network models also proved to be good tools for detecting drivers' engagement in secondary tasks. A proposed framework of crash index calculation provides an insight into how the crash risk associated with distracted driving behavior can be quantified. Further research is required to identify the required statistical analysis for the crash index calculation as well as provide further details on how such index can be used.
-
Format:
-
Funding:
-
Collection(s):
-
Main Document Checksum:
-
Download URL:
-
File Type: