In this paper, we present work on automatic road extraction from high resolution aerial imagery taken over urban areas. In order to deal with the high complexity of this type of scenes, we integrate detailed knowledge about roads and their context using explicitly formulated scale-dependent models. The knowledge about how and when certain parts of the road and context model are optimally exploited is expressed by an extraction strategy. The key feature of the presented approach is the integral treatment of three essential issues of object extraction in complex scenes: 1) Specific parts of the road model and extraction strategy are automatically adapted to the respective contextual situation. 2) The extraction incorporates components for self-diagnosis that internally evaluate hypotheses indicating their relevance for further processing. 3) Multiple views on the scene are utilized in different ways. Redundancies in the extraction are exploited, occlusions are predicted and obviated, and a 3D object description is generated. The results achieved with our approach show that a stringent realization of these issues enables the extraction of roads even if their appearance is heavily affected by other objects. Based on an external evaluation of the results, we discuss advantages but also remaining deficiencies of this approach.

Today's computer vision applications often have to deal with multiple, uncertain, and incomplete visual information. In this paper, we introduce a new method, termed `active fusion', which provides a common framework for active selection and combination of information from multiple sources in order to arrive at a reliable result at reasonable costs. The implementation of active fusion on the basis of probability theory, the Dempster-Shafer theory of evidence and fuzzy sets is discussed. In a sample experiment, active fusion using Bayesian networks is applied to agricultural field classification from multitemporal Landsat imagery. This experiment shows a significant reduction of the number of information sources required for a reliable decision. Keywords: information fusion, image understanding, active fusion, probability theory, Bayesian networks, Dempster-Shafer theory of evidence, fuzzy sets, fuzzy measures, entropy 1 Motivation Information fusion deals with the integration of info...

A scene analysis system for the 3--D modeling of objects is presented. It combines surface reconstruction techniques with object recognition for the generation of 3--D models for computer graphic applications. The system permits the insertion of highlevel constraints, like a specific angle between two house walls, in an explicit knowledge base implemented as a semantic net. The applicability of those constraints is proved by asserting and testing hypotheses in an interpretation phase. In the case of rejection a more general constraint or model is selected. The capabilities of the system were shown for the modeling of buildings using depth from stereo and contour information. The system reconstructs the surface of the scene objects using the constraints selected in the prior interpretation. 1 Introduction For the visualization of virtual environments, for example in applications like flight and driving simulators, architecture and landscape planning, there exists a high demand for photo...

Because of the increasing amount of remotely sensed imagery there is a growing need for efficient data analysis techniques. Here to automate the image interpretation a knowledge based approach is suggested. The presented

After a brief overview of the research going on in the Pastis group at Inria, which covers retrieving depth and recovering symbolic information from remotely sensed data, we concentrate in this paper on a particularly difficult task, high-resolution stereo, which deals with images with a resolution of one meter or less, on which there are smooth textured natural areas as well as discontinuities, or even occultations, of man-made structures. Two approaches are proposed for extracting reliable dense depth maps: one relies on the use of several aerial images, with increasing disparities, and the other on an adaptive window correlation scheme, which prevents the correlation window to extend over radiometric, and thus depth discontinuities.

mage fusion refers to the acquisition, processing and synergistic combination of information provided by various sensors or by the same sensor in many measuring contexts. The aim of this survey paper is to describe three typical applications of data fusion in remote sensing. The first study case considers the problem of the Synthetic Aperture Radar (SAR) Interferometry, where a pair of antennas are used to obtain an elevation map of the observed scene; the second one refers to the fusion of multisensor and multitemporal (Landsat Thematic Mapper and SAR) images of the same site acquired at different times, by using neural networks; the third one presents a processor to fuse multifrequency, multipolarization and mutiresolution SAR images, based on wavelet transform and multiscale Kalman filter. Each study case presents also results achieved by the proposed techniques applied to real data.

This paper presents a scene analysis system, called MESSIE, in a 3D robotic context. It is a blackboard based multispecialist system. The basic architecture of MESSIE has been extended to meet the requirements of a vision reasoning system for a robotic platform and to generalize the application domain of our previous works [7, 4, 3, 5]. In the first part, we present the system architecture and the models of objects, scene and sensors and the different interpretation and detection strategies. After, we present current experiments on indoor scene interpretation and an interpretation running example using constrained low-level feature extraction mechanism to improve low-level processing results.

A knowledge based approach for the interpretation of aerial images is presented that combines cues from multiple sensors (visual, infrared, SAR). Here the application of road extraction is described in detail. The sensor fusion is applied at object level. This allows to use prior knowledge to increase the separability of the classes. The prior knowledge is represented explicitly using semantic nets. Interpretation exploits the semantic net to control the sequence of sensor fusion mixing bottom up and top down strategies. The presented approach addresses the problem of uncertain and imprecise sensor data by judging the different cues based on possibility theory. Competing interpretations are stored in a search tree. An A* algorithm selects the most promising, i.e. best judged, interpretation for further investigation. Results are shown for the detection of roads in urban and agricultural areas exploiting image data from multiple sensors. KURZFASSUNG Im vorliegenden Beitrag wird ein wissensbasierter Ansatz zur Interpretation von Luftbildern vorgestellt, der Merkmale aus unterschiedlichen Sensordaten (visuell, IR, SAR) verarbeiten kann. Dieses wird am Beispiel der automatischen Extraktion von StraSen demonstriert. Die Sensor Fusion wird auf Objekt Ebene durchgetihrt, wodurch es m6glich wird, Vorwissen zu nutzen, um die Separierbarkeit der Objektklassen zu erh6hen. Das Vorwissen wird explizit in einem semantischen Netz repriisentiert, was wiihrend der Interpretation zur Steuerung der Sensr Fusin genutzt wird' Dabei werden abwechselnd daten und modellgetriebene Strategien eingesetzt. Die Unsicher_ heir und Ungenauigkeit der Daten werden durch ein Bewertungskalktil auf Basis der M6glichkeitstheorie berticksichtigt. Konkurrierende Interpretationen werden in einem Suchbaum abg...

The automated generation of 3D CAD models of real objects from different camera views poses frequently problems in regard to man made objects. Models do not match the expectations of a human observer, because house walls are not perpendicular, streets are not planar, windows and doors are not rectangular, etc.. The new knowledge based modeling system AIDA handles these problems by using an explicit knowledge base about the semantics of the scene to be modeled including knowledge about the visual appearance of scene objects. During the analysis of the scene constraints for the modeling are derived automatically and are applied during model generation. Keywords: Image Processing, Scene Analysis, Knowledge based System, 3--D Modeling, CAD Models, Virtual Reality 1 Introduction Presently a great demand for highly realistic looking 3--D models of real objects can be observed.These models are used in developing flight and driving simulators, in the movie and TV production, in advertising,...

This paper deals with the integration of domain knowledge to improve the landcover classification of a sequence of images. This new approach consists in representing the plot of land as a dynamic system and in modeling its evolution (knowledge about crop cycles, rotations and farmer practices) with the timed automata formalism. The main feature of this work is to improve the classification provided by a traditional classification with data resulting from the simulation of the plot evolution model. The aim of this paper is to focus on the experiments carried out on a sequence of five images. The problem of classification refinement and the model used to capture domain knowledge are first presented. The emphasis is then put on the results and their interpretation that show the contribution of the method to improve the classification of images