Discovering users' behavior via eye-tracking data analysis is a common task that has important implications in many domains including marketing, design, behavior study, and psychology. In our project, we are interested in analyzing eye-tracking data to investigate differences between age groups in emotion regulation using visual attention. To achieve this goal, we adopted a general-purposed interactive visualization method, namely Glyph, to conduct temporal analysis on participants' fixation data. Glyph facilitates comparison of abstract data sequences to understand group and individual patterns. In this article, we show how a visualization system adopting the Glyph method can be constructed, allowing us to understand how users shift their fixations and dwelling given different stimuli, and how different user groups differ in terms of these temporal eye-tracking patterns. The discussion demonstrates the utility of Glyph not only for the purpose of our project, but also for other eye-tracking data analyses that require exploration within the space of temporal patterns.
This paper presents a method to build a saliency map in a volumetric dataset using 3D eye tracking. Our approach acquires the saliency information from multiple views of a 3D dataset with an eye tracker and constructs the 3D saliency volume from the gathered 2D saliency information using a tomographic reconstruction algorithm. Our experiments, on a number of datasets, show the effectiveness of our approach in identifying salient 3D features that attract user’s attention. The obtained 3D saliency volume provides importance information and can be used in various applications such as illustrative visualization.
We present two visualization approaches illustrating the value of formal cognitive models for predicting, capturing, and understanding eye tracking as a manifestation of underlying cognitive processes and strategies. Computational cognitive models are formal theories of cognition which can provide predictions for human eye movements in visual decision-making tasks. Visualizing the internal dynamics of a model provides insights into how the interplay of cognitive mechanisms influences the observable eye movements. Animation of those model behaviors in virtual human agents gives explicit, high fidelity visualizations of model behavior, providing the analyst with an understanding of the simulated human’s behavior. Both can be compared to human data for insight about cognitive mechanisms engaged in visual tasks and how eye movements are affected by changes in internal cognitive strategies, external interface properties, and task demands. We illustrate the visualizations on two models of visual multitasking and juxtapose model performance against a human operator performing the same task.
Eye movement based analysis is becoming ever prevalent across domains with the commoditization of eye-tracking hardware. Eye-tracking datasets are, however, often complex and difficult to interpret and map to higher-level visual-cognitive behavior. Practitioners using eye tracking need tools to explore, characterize and quantify patterned structures in eye movements. In this paper, we introduce the VERP (Visualization of Eye movements with Recurrence Plots) Explorer, an interactive visual analysis tool for exploring eye movements during visual-cognitive tasks. The VERP Explorer VERP Explorer couples conventional visualizations of eye movements with recurrence plots recurrence plots that reveal patterns of revisitation over time. We apply the VERP Explorer to the domain of medical checklist design checklist design , analyzing eye movements of doctors searching for information in checklists under time pressure.
The relationship between where people look and where people reach has been studied since the dawn of experimental psychology. This relationship has implications for the designs of interactive visualizations, particularly for applications involving touchscreens. We present a new visual-motor visual-motor analytics dashboard for the joint study of eye movement and hand/finger movement movement dynamics dynamics. Our modular approach combines real-time playback of gaze and finger-dragging behavior together with statistical models quantifying the dynamics of both modalities. To aid in visualization and inference with these data, we apply Gaussian process Gaussian process regression models which capture the similarities and differences between eye and finger movements, while providing a statistical model of the observed functional data. Smooth estimates of the dynamics are included in the dashboard to enable visual-analytic exploration of visual-motor behaviors on touchscreen interfaces.
Location is an important component of a narrative. Mapped place names provide vital geographical, economic, historical, political, and cultural context for the text. Online sources such as news articles, travel logs, and blogs frequently refer to geographic locations, but often these are not mapped. When a map is provided, the reader is still responsible for matching references in the text with map positions. As they read a place name within the text, readers must locate its map position, then find their place again in the text to resume reading, and repeat this for each toponym. We propose a gaze-based reading and dynamic geographic information system (GazeGIS) which uses eye tracking and geoparsing to enable a more cohesive reading experience by dynamically mapping locations just as they are encountered within the text. We developed a prototype GazeGIS application and demonstrated its application to several narrative passages. We conducted a study in which participants read text passages using the system and evaluated their experience. We also explored an application for intelligence analysis and discuss how experts in this domain envision its use. Layman and intelligence expert evaluations indicate a positive reception for this new reading paradigm. This could change the way we read online news and e-books, the way school children study political science and geography, the way officers study military history, the way intelligence analysts consume reports, and the way we plan our next vacation.
The visual analysis of eye movement data has become an emerging field of research leading to many new visualization techniques in recent years. These techniques provide insight beyond what is facilitated by traditional attention maps and gaze plots, providing important means to support statistical analysis and hypothesis building. There is no single “all-in-one” visualization to solve all possible analysis tasks. In fact, the appropriate choice of a visualization technique depends on the type of data and analysis task. We provide a taxonomy of analysis tasks that is derived from literature research of visualization techniques and embedded in our pipeline model of eye-tracking visualization. Our task taxonomy is linked to references to representative visualization techniques and, therefore, it is a basis for choosing appropriate methods of visual analysis. We also elaborate on how far statistical analysis with eye-tracking metrics can be enriched by suitable visualization and visual analytics visual analytics techniques to improve the extraction of knowledge during the analysis process.
An observer’s eye movements are often informative about how the observer interacts with and processes a visual stimulus. Here, we are specifically interested in what eye movements reveal about how the content of information visualizations is processed. Conversely, by pooling over many observers’ worth of eye movements, what can we learn about the general effectiveness of different visualizations and the underlying design principles employed? The contribution of this manuscript is to consider these questions at a large data scale, with thousands of eye fixations on hundreds of diverse information visualizations. We survey existing methods and metrics for collective eye movement analysis, and consider what each can tell us about the overall effectiveness of different information visualizations and designs at this large data scale.
Using coefficient 𝒦 , defined on a parametric scale, derived from processing a traditionally eye-tracked time course of eye movements, we propose a straightforward method of visualizing ambient/focal fixations in both scanpath and heatmap visualizations. The 𝒦 coefficient indicates the difference of fixation duration and following saccade amplitude expressed in standard deviation units, facilitating parametric statistical testing. Positive and negative ordinates of 𝒦 indicate focal or ambient fixations, respectively, and are colored by luminance variation depicting relative intensity of focal fixation.
Eye-movements are typically measured with video cameras and image recognition algorithms. Unfortunately, these systems are susceptible to changes in illumination during measurements. Electrooculography (EOG) is another approach for measuring eye-movements that does not suffer from the same weakness. Here, we introduce and compare two methods that allow us to extract the dwells of our participants from EOG signals under presentation conditions that are too difficult for optical eye tracking. The first method is unsupervised and utilizes density-based clustering. The second method combines the optical eye-tracker’s methods to determine fixations and saccades with unsupervised clustering. Our results show that EOG can serve as a sufficiently precise and robust substitute for optical eye tracking, especially in studies with changing lighting conditions. Moreover, EOG can be recorded alongside electroencephalography (EEG) without additional effort.
Many applications such as data visualization or object recognition benefit from accurate knowledge of where a person is looking at. We present a system for accurately tracking gaze positions on a three dimensional object using a monocular head mounted eye tracker. We accomplish this by (1) using digital manufacturing to create stimuli whose geometry is know to high accuracy, (2) embedding fiducial markers into the manufactured objects to reliably estimate the rigid transformation of the object, and, (3) using a perspective model to relate pupil positions to 3D locations. This combination enables the efficient and accurate computation of gaze position on an object from measured pupil positions. We validate the accuracy of our system experimentally, achieving an angular resolution of 0.8∘ and a 1.5
In contrast to traditional video, immersive video allows viewers to interactively control their field of view in a 360∘ panoramic scene. However, established methods for the comparative evaluation of gaze data for video require that all participants observe the same viewing area. We therefore propose new specialized visualizations and a novel visual analytics framework for the combined analysis of head movement and gaze data. A novel View Similarity visualization highlights viewing areas branching and joining over time, while three additional visualizations provide global and spatial context. These new visualizations, along with established gaze evaluation techniques, allow analysts to investigate the storytelling of immersive videos. We demonstrate the usefulness of our approach using head movement and gaze data recorded for both amateur panoramic videos, as well as professionally composited immersive videos.
In user studies, eye tracking is often used in combination with other recordings, such as think-aloud protocols. However, it is difficult to analyze the eye-tracking data and transcribed recordings together because of missing data alignment and integration. We suggest the use of word-sized eye-tracking visualizations to augment the transcript with important events that occurred concurrently to the transcribed activities. We explore the design space of such graphics by discussing how existing eye-tracking visualizations can be scaled down to word size. The suggested visualizations can optionally be combined with other event-based data such as interaction logs. We demonstrate our concept by a prototypical analysis tool.
Analysis and visualization of eye movement data from eye-tracking studies typically take into account gazes, fixations, and saccades of both eyes filtered and fused into a combined eye. Although this is a valid strategy, we argue that it is also worth investigating low-level eye-tracking data prior to high-level analysis, because today’s eye-tracking systems measure and infer data from both eyes separately. In this work, we present an approach that supports visual analysis and cleansing of low-level time-varying data for eye-tracking experiments. The visualization helps researchers get insights into the quality of the data in terms of its uncertainty, or reliability. We discuss uncertainty originating from eye tracking, and how to reveal it for visualization, using a comparative approach for disagreement between plots, and a density-based approach for accuracy in volume rendering. Finally, we illustrate the usefulness of our approach by applying it to eye movement data recorded with two state-of-the-art eye trackers.
Eye tracking metrics may provide unobtrusive measures of cognitive states such as workload and fatigue and can serve as useful inputs into future human computer interface technologies. To further explore the usefulness of eye tracking for the estimation of cognitive state, the current experiment evaluated saccade, fixation, and pupil-based measures to identify which metrics reliably indexed cognitive workload in a dynamic, unconstrained task (Tetris ® ). In line with previous studies, our results show that some eye movement features are correlated with changes in workload, manipulated here via task difficulty. Among these were blink duration, saccade velocity, and tonic pupil dilation.
This paper describes the use of the Levenshtein distance and nearest neighbour index to visualise and analyse differences in eye-tracking scanpaths applied to the field of electrocardiology. Data was obtained from clinicians as they interpreted 12-lead electrocardiograms (ECGs). The main aim of the work is provide methods of visualising the differences between multiple participants scanpaths simultaneously. Allowing us to answer questions such as, do clinicians fixate randomly on the ECG, or do they apply a systematic approach? Results indicate that practitioners have very different search strategies applied to the majority of stimuli. The distribution of fixations is not random and tends towards clustering with all stimuli. The differences between practitioners are likely to be the result of different training, clinical role and expertise.
Word-sized visualizations for eye movement data allow analysts to compare a variety of experiment conditions or participants at the same time. We implemented a set of such word-sized visualizations as part of an analysis framework. We want to find out which of the visualizations is most suitable for different analysis tasks. To this end, we applied the framework to data from an eye tracking study on the reading behavior of users studying metro maps. In an expert evaluation with five analysts, we identified distinguishing characteristics of the different word-sized visualizations.
Attention maps-often in the form of heatmaps-are a common visualization approach to obtaining an overview of the spatial distribution of gaze data from eye tracking experiments. However, attention maps are not designed to let us easily analyze the temporal information of gaze data: they completely ignore temporal information by aggregating over time, or they use animation to build a sequence of attention maps. To overcome this issue, we introduce Hilbert attention maps: a 2D static visualization of the spatiotemporal distribution of gaze points. The visualization is based on the projection of the 2D spatial domain onto a space-filling Hilbert curve that is used as one axis of our new attention map; the other axis represents time. We visualize Hilbert attention maps either as dot displays or heatmaps. This 2D visualization works for data from individual participants or large groups of participants, it supports static and dynamic stimuli alike, and it does not require any preprocessing or definition of areas of interest. We demonstrate how our visualization allows analysts to identify spatiotemporal patterns of visual reading behavior, including attentional synchrony and smooth pursuit.
Nowadays, the amount of gaze data records of subjects associated with video sequences increases daily. These eye tracking data are unfortunately stored in separate files in custom-made data formats, which reduces accessibility even for experts and makes the data effectively inaccessible for non-experts. Consequently, we still lack interfaces for many common use cases, such as visualization, streaming, data analysis, high level understanding, and semantic web integration of eye tracking data. To overcome these shortcomings, we want to promote the use of existing multimedia container formats to establish a standardized method of incorporating content videos with eye tracking metadata. This will facilitate instantaneous visualization in standard multimedia players, streaming via the Internet, and easy usage without conversion. Using our prototype software, we embed gaze data from eye tracking studies and the corresponding video into a single multimedia container, which can be visualized by any media player. Based on this prototype implementation, we discuss the benefit of our approach as a possible standard for storing eye tracking metadata including the corresponding video.
We here present parts of our ongoing work to facilitate the largescale analysis of smooth pursuit eye movements made while viewing dynamic natural scenes. Classification of smooth pursuit episodes can be difficult in the presence of eye-tracking noise, and we thus recently proposed an algorithm that clusters gaze recordings from several observers in order to improve classification robustness. We now implemented a publicly available tool that allows for generation of a ground truth benchmark by assisted handlabelling of video gaze data. Based on the labelling produced with the tool we present preliminary evaluation results for our smooth pursuit classification approach in comparison to state-of-the-art algorithms. Overall, human observers spend more than 12% of their viewing time performing smooth pursuit, which emphasizes the importance of investigating smooth pursuit behaviour in naturalistic contexts.
