Roadway and Work Crew Conspicuity

 

Joseph E. Barton and James A. Misener, PATH

A state-of-the-art visual signature analysis tool that could be used to measure and improve the conspicuity of human and roadway hazards on California highways could serve as a powerful, cost-effective and semi-automated method of "virtual prototyping" in which drivers' perception of increased conspicuousness could be gauged. Notional designs and configurations could be simulated with very little investment, under different geometries, color/illumination combinations and ambient lighting conditions. The "best" design which optimizes some combination of cost-effectiveness and safety could then be confidently built and implemented. Our work quantitatively addresses the measurement of conspicuity of highway features and Caltrans work zones, from the perspective of driver detection.

The method focused on acquiring and operating on a computational visual signature analysis tool, but it evolved into evaluating the detection process, then selecting and exercising human perception-acquisition models suitable for development into a tool for conspicuity measurement. A composite, quantitative model of conspicuity was developed, verified, and applied to some sample roadside scenes. The technique developed was an objective, quantitative algorithm centered on contrast sensitivity, a known factor in conspicuity. The algorithm was tested using a still image of a roadside scene in which a pedestrian was wearing a Caltrans safety jacket.

Additional tests were conducted using a mathematically derived pattern with a region of high conspicuity and two versions of the original roadside scene that had been adjusted to demonstrate more and less conspicuity. In all cases, the algorithm provided a scoring grid which highlighted the regions of high conspicuity, thus demonstrating that the algorithm is valid. Therefore, this technique provides an objective and quantitative method for testing the conspicuity of roadside scenes. While additional modifications to the algorithm would provide increased precision, the work here provides an excellent first pass for those interested in testing the conspicuity of roadside scenes as well as a method for examining how changes in the scene affect conspicuity. The fundamental aspect of human vision that we exploit in this study is Contrast Sensitivity. The human visual system is tuned to detect not the absolute luminance levels of various areas in a visual scene, but rather the contrast between the luminance levels of adjacent areas. The signal supplied by the eye to the central visual structures does not give equal weight to all regions of the visual scene. Rather, it emphasizes the regions that contain the most information, namely, the regions where there are differences in luminance. Such regions are where our attention is directed when we look out into the world: they are the most conspicuous. Contrast sensitivity, then, underlies our ability to detect objects of interest in the visual scene, and also to discern patterns. This capability is of course the result of the way in which the human visual system has evolved. Appendix I, The Anatomy of Human Sight, describes the elements and structure of the human visual system that give rise to contrast sensitivity. The nature of light itself has significantly influenced this evolution, and a complete understanding of contrast sensitivity requires in turn an understanding of the Physics of Sight, which is taken up in Appendix II.

Full Report:http://www.path.berkeley.edu/publications/pdf/prr 2000/prr-2000-23.pdf