Abstract
Search in the real-world is often carried out in highly cluttered environments. Although several studies have explored the relationship between search and scene clutter, this effort has been hampered by a lack of stimuli varying systematically along the clutter dimension. We had observers search through quasi-realistic scenes created with the game Sim City, with each scene representing a different time in a city's evolution. The target in each scene was a visually designated building that was always present. The search task terminated when the target was fixated and confirmed with a button-press. There were three cities (rural, suburban, and urban), with 30 screenshots obtained for each scene type. Importantly, these 30 images captured a given city as it grew over time within the game (e.g., urban image 1 was an open field with some roads; urban image 30 was a bustling metropolis). Both between and within scene estimates of clutter were validated using independent raters, allowing us to examine how systematic changes in clutter affect search. We found longer RTs and more fixations in the heavily cluttered urban city (∼7s) compared to the minimally cluttered rural city (∼4s). Search in the suburban city produced an intermediate result (∼5s). Within city analyses revealed a similar pattern. As cities grew and became more cluttered, search became more difficult, a relationship reflected in the slope of the RT × Time in Game (TIG) and fixation number × TIG functions. Moreover, slopes were steepest in the urban city and shallowest in the rural city, indicating that clutter levels evolved at different rates according to city type. These data are broadly consistent with the view that search efficiency declines as scene clutter increases, and suggests that our Sim City stimuli may serve as a benchmark dataset for the evaluation of models of visual clutter.
This work was supported by grants from the NIH (R01 MH63748) and ARO (DAAD19-03-1-0039).