Abstract
Some rear-end collisions occur due to misjudgments of following distances; thus, it is important that drivers detect approaching or looming objects. We wanted to learn what visual information drivers look for under such circumstances. The classification image technique is thought to reveal to what in an image an observer attends. We restricted our study to some of the monocular cues to depth. A computer displayed two scenes rapidly in succession. Observers were fully corrected. A filled square of 900 × 900 pixels with adjustable positive contrast was displayed for 0.1 sec. In the next 0.1 sec epoch the square could either remain unchanged or could expand by 15 pixels (as if looming). The subject's task was to decide on a given trial if the square had expanded or not. Contrast was adjusted as needed to keep performance near 75% correct. The classification image for each of three young adult fully corrected subjects was obtained for 10,000 trials cumulated over several sessions. Images differed slightly, but common trends were seen. Patterns that influenced the observer's decision included high luminance at the upper and lateral edges and corners of the square and low luminance in the surround. The lower edge of this figure had little influence.
Caltrans and California PATH, Guidant Foundation