Optical flow based head movement and gesture analysis in automotive environment

Abstract

Head gesture detection and analysis is a vital part of looking inside a vehicle when designing intelligent driver assistance systems. In this paper, we present a simpler and constrained version of Optical flow based Head Movement and Gesture Analyzer (OHMeGA) and evaluate on a dataset relevant to the automotive environment. OHMeGA is user-independent, robust to occlusions from eyewear or large spatial head turns and lighting conditions, simple to implement and setup, real-time and accurate. The intuitiveness behind OHMeGA is that it segments head gestures into head motion states and no-head motion states. This segmentation allows higher level semantic information such as fixation time and rate of head motion to be readily obtained. Performance evaluation of this approach is conducted under two settings: controlled in laboratory experiment and uncontrolled on-road experiment. Results show an average of 97.4% accuracy in motion states for in laboratory experiment and an average of 86% accuracy overall in on-road experiment.