In the Visual Attention Lab, we study the attentional mechanisms underlying human vision. We are
particularly interested in investigating how attention is controlled for efficient performance of visual
tasks. Our main research paradigms are eye-movement recording and computational data analysis,
including machine learning, modeling, and simulation. The findings resulting from this research can
also inform the construction of computer vision systems and human-computer interfaces.
More specifically, there are four main areas of research that we focus on:
Studying Human Eye Movements
We use SR Research EyeLink-2K (remote) and EyeLink-II (head-mounted) systems to record eye movements during visual tasks. Eye movements provide insight into attentional processes that form the basis for efficient conscious processing of visual information. We employ tasks such visual search, comparative visual search, scene inspection, and reading to investigate the interaction of visual attention, eye movements, and visual working memory and gain a better understanding of the visual system. Besides basic research on the visual system, we also conduct clinically oriented eye-movement research on schizophrenia, Parkinson's disease, low vision, and dyslexia. We develop and apply advanced methods of data analysis to extract a maxmum of relevant information from the recorded data.
Computational Modeling of Cognitive and Perceptual Processes
Our modeling and simulation of processes in the human brain is aimed at two goals: First, we would like to test whether our conclusions from empirical studies are plausible. For this purpose we build a computational model of the hypothesized processes in the visual system, and then simulate these processes on a computer. A comparison of simulated and actual human behavior (with regard to eye movement patterns, psychometric functions, etc.) can identify the strengths and weaknesses of the current model and guide its refinement. Iterating this process leads to a plausible model and insight into the underlying processes and their interaction. Second, we investigate whether our models can be used to improve technical systems such as computer vision applications, human-computer interfaces, or image retrieval systems.
Machine Learning for Data Analysis and Computer Vision
We explore, develop, and utilize machine learning approaches to enable higher-level functionality in technical systems. Such functions include object recognition, scene understanding, user identification, activity recognition, and complex behaviors. They cannot be modeled adaequately through "hard-coded" algorithms or systems based on explicit sets of rules. Machine learning not only enables such functionalities but also allows them to adapt to changing environments and situations. Furthermore, machine learning is an important tool for the analysis of large databases. For example, in order to learn from biological systems for the improvement of technical approaches, we have to extract the most relevant data from huge sets of empirically collected data. Machine learning approaches such as deep neural networks can accomplish such tasks and thereby help us bridge the gap between biological and technical systems.
Improving Human-Computer Interfaces
Deeper insight into the human visual system, including visual attention and working memory, allows the optimization of human-computer interfaces. If gaze information is available during interface use, it can even be used to make the interface "smarter," i.e., adapt to the user and infer the user's intent from eye movement patterns. Furthermore, eye tracking can be used as a very fast, effortless, and intuitive way of directly controlling a computer program. However, not all eye movements are control-related and completely under conscious command, which needs to be considered for the construction of gaze-controlled interfaces. In the Visual Attention Lab, we build gaze-controlled interfaces for physically challenged users as well as multimodal interfaces for specific control tasks performed by healthy operators. In order to improve gaze-controlled human-computer interaction, we study eye-hand timing issues such as the eye-hand span that affects the use of multimodal interfaces.
More specifically, there are four main areas of research that we focus on:
Studying Human Eye Movements
We use SR Research EyeLink-2K (remote) and EyeLink-II (head-mounted) systems to record eye movements during visual tasks. Eye movements provide insight into attentional processes that form the basis for efficient conscious processing of visual information. We employ tasks such visual search, comparative visual search, scene inspection, and reading to investigate the interaction of visual attention, eye movements, and visual working memory and gain a better understanding of the visual system. Besides basic research on the visual system, we also conduct clinically oriented eye-movement research on schizophrenia, Parkinson's disease, low vision, and dyslexia. We develop and apply advanced methods of data analysis to extract a maxmum of relevant information from the recorded data.
Computational Modeling of Cognitive and Perceptual Processes
Our modeling and simulation of processes in the human brain is aimed at two goals: First, we would like to test whether our conclusions from empirical studies are plausible. For this purpose we build a computational model of the hypothesized processes in the visual system, and then simulate these processes on a computer. A comparison of simulated and actual human behavior (with regard to eye movement patterns, psychometric functions, etc.) can identify the strengths and weaknesses of the current model and guide its refinement. Iterating this process leads to a plausible model and insight into the underlying processes and their interaction. Second, we investigate whether our models can be used to improve technical systems such as computer vision applications, human-computer interfaces, or image retrieval systems.
Machine Learning for Data Analysis and Computer Vision
We explore, develop, and utilize machine learning approaches to enable higher-level functionality in technical systems. Such functions include object recognition, scene understanding, user identification, activity recognition, and complex behaviors. They cannot be modeled adaequately through "hard-coded" algorithms or systems based on explicit sets of rules. Machine learning not only enables such functionalities but also allows them to adapt to changing environments and situations. Furthermore, machine learning is an important tool for the analysis of large databases. For example, in order to learn from biological systems for the improvement of technical approaches, we have to extract the most relevant data from huge sets of empirically collected data. Machine learning approaches such as deep neural networks can accomplish such tasks and thereby help us bridge the gap between biological and technical systems.
Improving Human-Computer Interfaces
Deeper insight into the human visual system, including visual attention and working memory, allows the optimization of human-computer interfaces. If gaze information is available during interface use, it can even be used to make the interface "smarter," i.e., adapt to the user and infer the user's intent from eye movement patterns. Furthermore, eye tracking can be used as a very fast, effortless, and intuitive way of directly controlling a computer program. However, not all eye movements are control-related and completely under conscious command, which needs to be considered for the construction of gaze-controlled interfaces. In the Visual Attention Lab, we build gaze-controlled interfaces for physically challenged users as well as multimodal interfaces for specific control tasks performed by healthy operators. In order to improve gaze-controlled human-computer interaction, we study eye-hand timing issues such as the eye-hand span that affects the use of multimodal interfaces.
If you are interested in conducting research in the Visual Attention Lab,
please don't hesitate to contact the Lab Director
please don't hesitate to contact the Lab Director
