Abstract
Perceiving the gist of real-world static images is fast. At VSS2018, Wolfe and Rosenholtz showed similar rapid understanding of videos of driving. On road, the driver’s environment changes quickly, requiring prompt hazard detection to avoid collisions, yet research in driving has argued that merely perceiving the environment requires seconds, not milliseconds. The Wolfe and Rosenholtz work demonstrated that drivers could detect pre-collision hazards in novel road scenes with as little as 200 ms of video. In addition, they showed that drivers could understand the scene sufficiently to evade the hazard with as little as 400 ms. However, these experiments did not constrain gaze. How might subjects’ ability to detect hazards change if they were forced to use peripheral vision exclusively, fixating a location similar to where Uber drivers mount their smartphones? Using stimuli from Wolfe and Rosenholtz’ 2018 work, drivers were required to maintain fixation at a location 20° below the center of the display and asked to detect hazards. This was done using a three down / one up staircase procedure to determine drivers’ individual video threshold durations for detecting hazards with peripheral vision alone. When forced to use peripheral vision, drivers’ mean thresholds rose significantly compared to free-viewing (787 ms in this experiment vs 468 ms in Wolfe and Rosenholtz’ 2018 work), a 319 ms penalty, equivalent to 10 m at 104 kph. So: why do Uber drivers scare you? They trust their peripheral vision too much. It is useful, but not sufficient to keep everyone safe.
Acknowledgement: TRI-CSAIL Joint Partnership