The effects of Visual Clutter and Perceptual Speed in high-pressured - - PDF document

UDT 2020 UDT Extended Abstract Template Presentation/Panel

The effects of Visual Clutter and Perceptual Speed in high-pressured information environments on the performance of tactical systems operators in the underwater battlespace

Olivia Foulds1, Dawn Wood2

1PhD Student, University of Strathclyde, Glasgow, Scotland 2 Engineer, BAE Systems Submarines, Frimley, England

Abstract — Tactical systems operators are required to work in high pressured, complex information environments where they are bombarded with a multitude of visual data. However when too much visual stimuli is present, the phenomenon of clutter degrades an individual’s perception in many tasks. This study presents an experiment that explored a user’s search behaviour, performance, and experience, given various manipulations of clutter depending on it’s congruence to the task. Furthermore, we investigated the interaction between clutter and perceptual speed, which is an individual difference for how accurately and quickly people process visual information. The exploration of these results will inform the design of future underwater battlespace applications to maximise the efficiency of human

• perators.

1 Background

In underwater battlespace, a vast amount of visual information, or clutter, is displayed to sonar operators. However, the phenomenon of ‘clutter’ occurs when the influence of nearby visual contours negatively interfere with and reduce visual perception when trying to focus

• n a target [1]. Clutter can therefore dramatically affect
• perator performance because the human brain has

limited attentional and perceptual capabilities and cannot process everything it sees. This is detrimental in an environment where an operator must make sense of what they see, while simultaneously maintaining situational awareness (SA) to act upon any events that may arise. Individuals vary in their perceptual ability to process visual information. The worse perception is, the more errors occur. Currently, tests for individual differences that infer how well operators respond to visual clutter and demonstrate SA are lacking. For example, SA tests make use of subjective ratings or require additional exercises to complete [2], which may not be reliable or detract an

• perator from their current task. Additionally, infrequent

tests do not account for the dynamic challenges that can affect perception, such as fatigue. Instead of finding ways to directly monitor individual differences, recent work has instead focused on developing advice to maintain factors such as aerobic fitness and hydration to ensure healthy cognitive performance [3]. However as military personnel face constant challenges and unpredictability, it would be beneficial if a system could dynamically understand the current ability of it’s operator, to ensure appropriate actions can occur if perception reduces. Perceptual Speed (PS) is a cognitive ability defined by an individual’s accuracy and speed to scan information while completing visual search tasks [4]. Although PS testing has been used in military contexts such as pilot selection, where the aim has been to employ higher PS scorers for jobs that require complex perceptual abilities [5], PS tests are old-fashioned and contain validity and reliability concerns [4]. Yet given the importance of understanding how an individual operator can process visual information, new PS tests may provide valuable

• insights. Thus, the present study designed two new

computerised versions of PS tests, to explore: a) how low and high users interact with various manipulations of visual clutter, and b) if it is possible to then predict PS based on search behaviour.

2 Approach

40 users completed two computerised PS tests and were categorised into low and high ability based on a median

• split. Each user then completed search tasks where they

were to find appropriate targets specific to each topic requirement amidst four variations of visual clutter: none, congruent, incongruent, and mixed. Congruence was chosen as an important clutter feature as previous work has identifed that although not necessary to complete a task successfully, viewing extra related information can increase system interaction and subsequently improve task accuracy [6]. Users also completed questionnaires. The effects of clutter and PS were then explored for a user’s search behaviour, performance, and experience. Finally, various machine learning models were trained based on search log data, to see if we could accurately predict PS without the need of administering a separate test.

3 Results

Similar to previous literature, clutter overall resulted in users: taking significantly longer to identify targets (p =

0.0067); performing significantly worse in a post-task

recall test; and reporting the most confusion and tiredness. Although congruent clutter increased interaction and more items were identified as targets,

• verall target accuracy did not improve.

Low PS users (which were 30% of our sample, and categorised as those who had scored low on both PS

UDT 2020 UDT Extended Abstract Template Presentation/Panel tests) were most negatively affected by incongruent clutter, showing their poorest performance in this

• condition. However, regardless of whether clutter was

present or not. low PS always clicked on and saved less targets, spent less time interacting with the task, and had worse accuracy compared to high PS. Yet, subjective ratings for confusion and tiredness did not differ between PS users. With such differences in behaviour between low and high PS users, a Decision Tree Machine Learning Model was then able to predict whether a user had low or high PS with up to 86% accuracy, based on a user completing just

• ne search. Additionally, our model demonstrated that

different search behaviours were important for different PS tests. Eg. only one PS test model was driven by time features, where low users spent longer making a decision.

4 Discussion

Although we predicted that congruent clutter would increase user interaction and subsequently improve performance [8], our results found that clutter of any kind causes negative effects on a user’s search performance and subjective feelings. This suggests that regardless of the content, if it visually has to be processed, performance degrades. This reaffirms work that demonstrated how peripherally viewed distracting objects can overload cognition even without direct gaze, leading to a bottleneck that impairs object perception [1]. In an environment that is heavily cluttered with sensor, navigational, and communication systems, our results would imply that even if one interface was not relevant to an operator’s current task, just being able to see it’s presence may be enough to distract them. We also demonstrated that low PS users are most negatively affected by clutter. This is important for the underwater battlespace industry, where an inability to correctly identify a target could result in a life or death

• scenario. Similar to the airforce employing PS screening

tests [6], we therefore propose that tactical systems

• perators should also undergo PS testing, in order to

select the best candidate with the highest level of visual

• abilities. However, there are many different types of PS

test available, and it is currently unclear how valid

• riginal tests are [5]. Yet, as our new PS tests identified

that they both measure different aspects, where one in particular identified users who struggled completing the task in an efficient time-scale, this would appear a more appropriate test to use for selecting operators who work in a situation where time is of the essence. However, individuals who generally perceive information well are still exposed to extreme, unpredictable environments in underwater battlespace. Consequently, environmental factors such as mental and physical fatigue can reduce an individual’s awareness

• f

their surroundings. We therefore believe that training individuals to support high performance standards [4] is not enough, and instead operating systems should be able to infer a user’s PS score to: a) produce a warning that a different operator should take over; or b) collaborate with agents that automatically adapt the visual interface to accommodate for the current lower PS. As we have also been the first experiment to show that PS can be predicted through a user’s behaviour with a system, with up to 86% accuracy, this initiates the first step into designing these real-time dynamic systems that will

• ptimally identify an operator who is struggling and

provide appropriate accommodations to help them when too much clutter is present. However before real-life application can occur, more research is needed to improve the model accuracy closer to 100%, and apply it directly in different contexts. In conclusion, we have added to the literature on the detrimental affects

• f

clutter, and in particular demonstrated that people who have lower levels of Perceptual Speed are most negatively affected in terms of search performance and experience. This reaffirms the importance for developing interfaces that exhibit minimal levels of clutter and dynamically adapt to infer and accommodate low PS users.

Acknowledgements

Thanks go to Dr Leif Azzopardi and Dr Martin Halvey for their supervision of the undertaken project discussed. This work was part funded by BAE Systems Maritime and EPSRC (EP/S513908/1).

References

[1] DM.Levi, Vision Res, 48:5, 635-654 (2008) [2] S.Loft, V.Bowden, J.Braithwaite, DB.Morrell, S.Huf, FT.Durso, HumFactors, 57:2, 298-310 (2013) [3] K.Martin, J.Periard, B.Rattray, DB.Pyne, Hum Factors, 62:1, 93-123 (2020) [4] O. Foulds, L. Azzopardi, M. Halvey, CHIIR (2020) [5] H.J.Hoermann, D.L.Damos, ISOAP, 85, 391-396 (2019) [6] J.Arguello, R.Capra, CIKM, 21, 1293-1302 (2012)

Author/Speaker Biographies

Olivia Foulds gained a First-Class Honours in Psychology at The Open University. Olivia is now a Computer and Information Science PhD student at The University of Strathclyde, working under a BAE Systems

• studentship. Dr Dawn Wood has a doctorate in

Chemistry, and uses her analytical skills in the

• perational analysis and technology research space as

part of BAE Systems. Whilst quite new to the defence industry, Dawn has 4+ years of data analysis experience from her academic background.

Public Access: Freedom of Information Act 2000 This document contains sensitive information as of the date provided to the original recipient by BAE Systems and the University of Strathclyde and is provided in

UDT 2020 UDT Extended Abstract Template Presentation/Panel

confidence and which may be subject to exclusions from access request under various sections of the legislation. Following a request for this information, the public authority shall provide BAE Systems and the University of Strathclyde with a copy of the request as soon as practicable following receipt, and shall afford BAE Systems and the University of Strathclyde the opportunity to make written representations in relation to any such request with reasonable notice. BAE Systems and the University of Strathclyde shall also be given prior written notice of any proposed disclosure by the public authority

• f any of the information contained in this document, or of a summary of any of this

information, whether in response to a freedom of information request or otherwise. Release of this information by a public authority without the prior written consent of BAE Systems and the University of Strathclyde may constitute an actionable breach of confidence