Kerstin Schill

Object and scene analysis by saccadic eye-movements : an investigation with higher-order statistics

Based on an information theoretical approach, we investigate feature selection processes in saccadic object and scene analysis. Saccadic eye movements of human observers are recorded for a variety of natural and artificial test images. These experimental data are used for a statistical evaluation of the fixated image regions. Analysis of second-order statistics indicates that regions with higher spatial variance have a higher probability to be fixated, but no significant differences beyond these variance effects could be found at the level of power spectra. By contrast, an investigation with higher-order statistics, as reflected in the bispectral density, yielded clear structural differences between the image regions selected by saccadic eye movements as opposed to regions selected by a random process. These results indicate that nonredundant, intrinsically two-dimensional image features like curved lines and edges, occlusions, isolated spots, etc. play an important role in the saccadic selection process which must be integrated with top-down knowledge to fully predict object and scene analysis by human observers.

Spatial Cognition: Reasoning, Action, Interaction

The Transregional Collaborative Research Center SFB/TR 8 Spatial Cognition pursues interdisciplinary research on a broad range of topics related to the representation and processing mechanisms for intelligent spatial behavior in technical and in natural systems. This contribution gives an overview of the field of research worked on in the SFB/TR 8 Spatial Cognition and presents three representative examples that illustrate the activities in the three research areas Reasoning, Action, and Interaction.

Scene analysis with saccadic eye movements: top-down and bottom-up modeling

The perception of an image by a human observer is usually modeled as a parallel process in which all parts of the image are treated more or less equivalently, but in reality the analysis of scenes is a highly selective procedure, in which only a small subset of image locations is processed by the precise and efficient neural machinery of foveal vision. To understand the principles behind this selection of the ‘‘informative’’ regions of images, we have developed a hybrid system that consists of a combination of a knowledgebased reasoning system with a low-level preprocessing by linear and nonlinear neural operators. This hybrid system is intended as a first step towards a complete model of the sensorimotor system of saccadic scene analysis. In the analysis of a scene, the system calculates in each step which eye movement has to be made to reach a maximum of information about the scene. The possible information gain is calculated by means of a parallel strategy which is suitable for adaptive reasoning. The output of the system is a fixation sequence, and finally, a hypothesis about the scene.

The “Intelligent Container”—A Cognitive Sensor Network for Transport Management

The “Intelligent Container” is a sensor network used for the management of logistic processes, especially for perishable goods such as fruit and vegetables. The system measures relevant parameters such as temperature and humidity. The concept of “cognitive systems” provides an adequate description of the complex supervision tasks and sensor data handling. The cognitive system can make use of several algorithms in order to estimate temperature related quality losses, detect malfunctioning sensors, and to control the sensor density and measurement intervals. Based on sensor data, knowledge about the goods, their history and the context, decentralized decision making is realized by decision support tools. The amount of communication between the container and the headquarters of the logistic company is reduced, while at the same time the quality of the process control is enhanced. The system is also capable of self-evaluation using plausibility checking of the sensor data.

KI 2006: Advances in Artificial Intelligence

Session 1. Invited Talk.- Expressivity-Preserving Tempo Transformation for Music - A Case-Based Approach.- Session 2. Cognition and Emotion.- MicroPsi: Contributions to a Broad Architecture of Cognition.- Affective Cognitive Modeling for Autonomous Agents Based on Scherers Emotion Theory.- Session 3A. Semantic Web.- OWL and Qualitative Reasoning Models.- Techniques for Fast Query Relaxation in Content-Based Recommender Systems.- Session 3B. Analogy.- Solving Proportional Analogies by E-Generalization.- Building Robots with Analogy-Based Anticipation.- Session 4A. Natural Language.- Classification of Skewed and Homogenous Document Corpora with Class-Based and Corpus-Based Keywords.- Learning an Ensemble of Semantic Parsers for Building Dialog-Based Natural Language Interfaces.- Session 4B. Reasoning.- Game-Theoretic Agent Programming in Golog Under Partial Observability.- Finding Models for Blocked 3-SAT Problems in Linear Time by Systematical Refinement of a Sub-model.- Towards the Computation of Stable Probabilistic Model Semantics.- DiaWOz-II - A Tool for Wizard-of-Oz Experiments in Mathematics.- Session 5. Invited Talk.- Applications of Automated Reasoning.- Session 6A. Ontologies.- On the Scalability of Description Logic Instance Retrieval.- Relation Instantiation for Ontology Population Using the Web.- Session 6B. Spatio-temporal Reasoning.- GeTS - A Specification Language for Geo-Temporal Notions.- Active Monte Carlo Recognition.- Session 7A. Machine Learning.- Cross System Personalization and Collaborative Filtering by Learning Manifold Alignments.- A Partitioning Method for Mixed Feature-Type Symbolic Data Using a Squared Euclidean Distance.- Session 7B. Spatial Reasoning.- On Generalizing Orientation Information in .- Towards the Visualisation of Shape Features The Scope Histogram.- Session 8A. Robot Learning.- A Robot Learns to Know People-First Contacts of a Robot.- Recombinant Rule Selection in Evolutionary Algorithm for Fuzzy Path Planner of Robot Soccer.- Session 8B. Classical AI Problems.- A Framework for Quasi-exact Optimization Using Relaxed Best-First Search.- Gray Box Robustness Testing of Rule Systems.- A Unifying Framework for Hybrid Planning and Scheduling.- Session 9. Agents.- A Hybrid Time Management Approach to Agent-Based Simulation.- Adaptive Multi-agent Programming in GTGolog.- Agent Logics as Program Logics: Grounding KARO.- On the Relationship Between Playing Rationally and Knowing How to Play: A Logical Account.- Special Event. 50 Years Artificial Intelligence.- 1956-1966 How Did It All Begin? - Issues Then and Now.- Fundamental Questions.- Towards the AI Summer.- History of AI in Germany and The Third Industrial Revolution.- Three Decades of Human Language Technology in Germany.- 1996-2006 Autonomous Robots.- Projects and Vision in Robotics.- What Will Happen in Algorithm Country?.

From Motion Observation to Qualitative Motion Representation

Since humans usually prefer to communicate in qualitative and not in quantitative categories, qualitative spatial representations are of great importance for user interfaces of systems that involve spatial tasks. Abstraction is the key for the generation of qualitative representations from observed data. This paper deals with the conversion of motion data into qualitative representations, and it presents a new generalization algorithm that abstracts from irrelevant details of a course of motion. In a further step of abstraction, the shape of a course of motion is used for qualitative representation. Our approach is motivated by findings of our own experimental research on the processing and representation of spatio-temporal information in the human visual system.

A model of visual spatio-temporal memory: the icon revisited.

With a minimal set of assumptions resulting from considerations about the perception of temporal structure, we argue for the existence of a spatio-temporal memory established by the mapping of time into simultaneous physical properties. The important point of this model is the distinction between external, physical time and the internal representation of time. An immediate consequence of such a structure is the emergence of properties usually associated with the concept of iconic memory or informational persistence. Some of these properties may hence be regarded as epiphenomena produced by the testing of a spatio-temporal system with tachistoscopic spatial stimuli. The model can explain properties of the immediate memory span, the lack of effect of exposure duration on tachistoscopic report, the partial-report superiority, the decay of iconic memory, and effects of a backward mask. It does not only avoid the incompatibility problems of the frozen-image concept in dynamic vision, but also provides an adequate basis for the processing of time-varying scenes.

Sensorimotor representation and knowledge-based reasoning for spatial exploration and localisation

We investigate a hybrid system for autonomous exploration and navigation, and implement it in a virtual mobile agent, which operates in virtual spatial environments. The system is based on several distinguishing properties. The representation is not map-like, but based on sensorimotor features, i.e. on combinations of sensory features and motor actions. The system has a hybrid architecture, which integrates a bottom-up processing of sensorimotor features with a top-down, knowledge-based reasoning strategy. This strategy selects the optimal motor action in each step according to the principle of maximum information gain. Two sensorimotor levels with different behavioural granularity are implemented, a macro-level, which controls the movements of the agent in space, and a micro-level, which controls its eye movements. At each level, the same type of hybrid architecture and the same principle of information gain are used for sensorimotor control. The localisation performance of the system is tested with large sets of virtual rooms containing different mixtures of unique and non-unique objects. The results demonstrate that the system efficiently performs those exploratory motor actions that yield a maximum amount of information about the current environment. Localisation is typically achieved within a few steps. Furthermore, the computational complexity of the underlying computations is limited, and the system is robust with respect to minor variations in the spatial environments.

Catching audiovisual mice: predicting the arrival time of auditory-visual motion signals.

We investigated the extent to which auditory and visual motion signals are combined when observers are asked to predict the location of a virtually moving target. In Condition 1, the unimodal and bimodal signals were noisy, but the target object was continuously visible and audible; in Condition 2, the virtually moving object was hidden (invisible and inaudible) for a short period prior to its arrival at the target location. Our main finding was that the facilitation due to simultaneous visual and auditory input is very different for the two conditions. When the target is continuously visible and audible (Condition 1), the bimodal performance is twice as good as the unimodal performances, thus suggesting a very effective integration mechanism. On the other hand, if the object is hidden for a short period (Condition 2) and the task therefore requires the extrapolation of motion speed over a temporal and spatial period, the facilitation due to both sensory inputs is almost absent, and the bimodal performance is limited by the visual performance.

Spatial Cognition VIII

A significant amount of research in robotics is aimed towards building robots that operate indoors yet there exists little analysis of how human spaces are organized. In this work we analyze the properties of indoor environments from a large annotated floorplan dataset. We analyze a corpus of 567 floors, 6426 spaces with 91 room types and 8446 connections between rooms corresponding to real places. We present a system that, given a partial graph, predicts the rest of the topology by building a model from this dataset. Our hypothesis is that indoor topologies consists of multiple smaller functional parts. We demonstrate the applicability of our approach with experimental results. We expect that our analysis paves the way for more data driven research on indoor