You are here

Computer Vision Measurements and Analysis to Support Naturalistic Driving Studies. Developing calibration and metrics for automated data extraction algorithm.

Project Information

Project ID: 
FHWA-PROJ-14-0051
Project Status: 
Completed
Start Date: 
Friday, June 28, 2013
End Date: 
Tuesday, May 31, 2016
FHWA Program: 
Exploratory Advanced Research
FHWA Topics: 
Safety--Data and Analysis Tools; Highway-Railroad Grade Crossing
TRT Terms: 
Human Factors; Research; Video; Data; Information Technology; Algorithms; Safety
FHWA Discipline: 
Safety
TRB Subject Area: 
Data and Information Technology, Safety and Human Factors, Research

Contact Information

First Name: 
Lincoln
Last Name: 
Cobb
Telephone: 
(202) 493-3313
Email Address: 
Team: 
Office of Safety Research and Development
Office:
Office of Research, Development, and Technology

Project Details

Project Description: 

Develop and apply the means to evaluate the automated video analysis algorithms currently being developed under EAR funding. Generalize those results so that researchers in various domains will have standard metrics and data sets available for their internal and external quality assurance/quality check (QA/QC) regimes.

Goals:

  • Develop processes and technology to evaluate the automated video analysis algorithms being developed by research teams funded by the Exploratory Advanced Research (EAR) program.
  • Create standard procedures and datasets, and performance metrics, which can be made available to the video analytics community.

Deliverables

Deliverable Name: 
Standard procedures, including standard datasets, for evaluating the performance of automated feature extraction, or automated identity masking algorithms.
Deliverable Type: 
Research report or guidelines
Deliverable Description: 
Support the development of tools that at least partially automate the process of coding large video datasets, such as the second Strategic Highway Research Program (SHRP2) Naturalistic Driving Studies (NDS) dataset (over 1 million hours of video data), or that automate the masking of the identity of NDS subjects visible in one or more video fields of view. In both cases, automated tools will dramatically open the pool of researchers able to apply large video data sets, such as the NDS, but reducing the time and the cost to do feature extraction or identity masking manually.