Photographic Data Extraction Feasibility and Pilot Study in Support of Roadside Safety and Roadway Departure Research

CHAPTER 1: INTRODUCTION

BACKGROUND

The research team undertook the photographic data extraction pilot study in response to the 2008 revision of the Fatality Analysis Reporting System (FARS), releasing geographical positioning system coordinates.⁽²⁾ Subsequently, the National Highway Traffic Safety Administration (NHTSA) released the geographical coordinates for previous crash years in an independent file format for 1999 and 2000 and revision of the FARS files from 2001 through 2007. Using the Federal Highway Administration (FHWA) roadway departure definition to filter crashes of interest, events involving guardrails, concrete barriers, and impact attenuators were reviewed from 2004 to 2009.⁽⁵⁾ From the pilot study, it was determined that barrier type and placement might provide useful information at low acquisition costs. The missing element, however, was post-crash condition. To obtain that information, it was necessary to review on-scene crash data. The current study considered the on-scene photography captured through the National Motor Vehicle Crash Causation Study (NMVCCS).⁽⁴⁾ NMVCCS was a congressionally mandated special study compiled under the National Automotive Sampling System (NASS) architecture from 2005 through 2007, with a primary objective to provide pre-crash and crash causation information.^(3,4)

In 2008, the NASS Crashworthiness Data System (CDS), providing a sample of tow-away crashes occurring on public roadways in the United States, showed marked improvement in roadside crash photography, possibly modeled after NMVCCS photography.⁽⁶⁾ Data collection for these crashes, however, continued to be near-crash time, with NASS researchers arriving to the crash scene within days or weeks. Prior NASS CDS efforts included the 1982 Longitudinal Barrier Special Study (LBSS), which was a collaborative effort with NHTSA.⁽⁷⁾ However, the photographs and crash data were filed in paper medium and electronic records unavailable to FHWA. To date, no NHTSA crashworthiness dataset has provided a suitable level of coded roadway environment data for FHWA analysis.

This study sought to complement the coded datasets by extracting new information from photographic images. NMVCCS data were selected owing to their temporal association with the crash and quality of photographs taken. Further, the pilot study initially attempted to classify the types of barriers involved, but the lack of temporal association limited the data extraction to an inventory confirming coded FARS attributes. The ability for researchers to assess the types of damage was considered useful as data inputs for future modeling tasks, thereby validating the immediate use of NMVCCS to feed the codification of a supplementary roadside element dataset.

National Cooperative Highway Research Program (NCHRP) projects have used photographic analysis to expand the understanding of uncoded elements, specifically NCHRP projects 17-11, 17-22, and 22-15.^(8–10) NCHRP projects 17-11 and 17-22 sought to identify impact conditions for run-off-road crashes, while project 22-15 considered the compatibility of the vehicle fleet with existing barrier types. The three studies compiled previously unavailable data from photographs and synthesized relevant test and real-world data.

Table 1 summarizes the means of harmonizing the coded and uncoded data. Historically, two broad types of data have existed: basic and crash. Basic data provided basic vehicle characteristics (e.g., make, model, and assessment of components) and occupant demography (e.g., observational and hospital discharge data). Crash data have generally provided post-crash vehicle disposition (e.g., tow away status and full deformation measurements) and geographical location (e.g., explicitly stated or sanitized scene photographs absent of geographical location information serving as a surrogate). Codified details have depended on the collection mandate of the dataset. It has been understood that in the absence of naturalistic data on each crash occurring, precise vehicle trajectory will be unavailable. However, the identification of roadside elements and descriptions of roadside environment have been available in crash photographs for several years. Currently, photographic data extraction is a labor-intensive task; however, in the future, roadside element extraction might benefit from automated data extraction based on the work of the FHWA Exploratory Advanced Research (EAR) Program.

The authors of this report advocate additional roadside element acquisition from the NMVCCS dataset. For FARS, a means of digitally extracting roadside element information from panoramic view software images is being assessed separately in tandem with findings of concurrent and related EAR projects. If NMVCCS and FARS data extraction methodologies are found to be acceptable for safety research, then a combination of these techniques will be applied to extract data from NASS CDS images.

OBJECTIVES

This study seeks to enhance the understanding of roadway element data from collected but uncodified information. This report seeks to perform the following tasks:

• Outline the findings of the photographic data extraction feasibility study.
  
  
• Describe the findings of the photographic data extraction pilot study to date.
  
  
• Provide recommendations for future photographic data extraction efforts.