Quantifying the effects of target and distractor similarity in complex visual search (STIR)
Abstract
1. Statement of scientific objective Visual search is comprised of several foundational cognitive tasks, including sensory perception, attention, and decision-making. Successful visual search is required across many situations, ranging from mundane (e.g., finding the milk in a well-stocked fridge) to life-or-death (e.g., warfighters screening for IEDs at roadside checkpoints). Given its importance, visual search has long been a topic of research interest (see Eckstein, 2011; Nakayama & Martini, 2011; and Wolfe, 2018 for recent reviews), with both basic and applied implications (e.g., Berbaum et al., 2013; Biggs, Kramer, & Mitroff, 2018; Biggs & Mitroff, 2015; Wetter, 2013;Wolfe, 2016). Due to limitations in statistical power, lab-based studies often use small sets of targets and distractors to quantify the impact of general factors on visual search (e.g., the number of distractors). However, real-world visual search often involves broad classes of targets among a large set of heterogeneous distractors; making it difficult for labbased studies to properly inform some of the more complex and critical aspects of real-world search. While several attempts have been made to increase the heterogeneity of stimuli and search dynamics (e.g., Kunar & Watson, 2011), time and other logistical constraints make it nearly impossible to achieve real-world-like conditions in the lab. 2. Methods to be employed Here we propose to use Òbig dataÓ married to computational vision analyses to characterize two critical aspects of the impact of the relationship between targets and distractors on behavior. First, we will utilize a pre-existing dataset gleaned from a smartphone app (Airport Scanner; https://www.airportscannergame.com) of over 3.75 billion trials of visual search to precisely quantify the efficiency of search to many different combinations of targets and distractors. Second, we will compare these results with multiple model-based and empiricallymeasured metrics of target-distractor similarity (e.g., convolutional neural net (CNN) layers, models of visual cortical areas, semantic priming) to establish which form(s) of similarity influence search performance. Finally, we will leverage the sheer volume of data to model how the effect of target-distractor similarity changes with training and experience. 3. Significance of the proposed effort to the advancement of knowledge The proposed research will extend our basic science understanding of visual search into a more complex and realistic regime by developing a new type of analysis and deploying it on the massive Airport Scanner behavioral dataset. The new knowledge produced will have implications for a multitude of areas of interest to the Army, including understanding the factors that affect human performance in any scenario in which the modernwarfighter is conducting a visual search in complex scenes (e.g., analyzing satellite defense images, detecting IEDs, searching for enemy combatants in a busy urban environment). Furthermore, a better understanding of the evolving effect of stimulus similarity on behavior holds the promise of improving training of visual search tasks both within and beyond the Army.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 31, 2020
- Source ID
- W911NF2010325
Entities
People
- Patrick Cox
Organizations
- Army Contracting Command
- George Washington University
- United States Army