Vehicle Crash Analysis Using human-in-the-loop Computer Vision & Crowdsourcing

Overview

Goal: Extraction of physical driving conditions and parameters from real, unconstrained crash videos with the help of a human in the loop.

Two-level framework:

  • First level: characterizes each video within predefined general types of conditions, called 'scenarios'.
  • Second level: associates quantitative information for each instance of scenario, called 'tokens'.
workflow


Contributions


I contributed to the project extensively by creating intelligent systems and running user studies for various purposes. My work are summarized in the following slides:





Go to top