This paper presents new techniques for improving the structural quality of automatically acquired architectural 3D models. Common architectural features like parallelism and orthogonality of walls and edges are exploited. The location of these features is extracted from the model by using a probabilistic technique (RANSAC). The relationships among the planes and edges are inferred automatically using a knowledge-based architectural model. A numerical algorithm is used to optimise the orientations of the features. Small irregularities in the model are removed by projecting the triangulation vertices onto the features. Planes and edges in the resulting model are aligned to each other. The techniques produce models with improved appearance. We show results for synthetic and real data with consideration of noise.

A knowledge based approach for the surface reconstruction of buildings to be used in computer graphic applications is presented. Using a calibrated stereo camera pair, scene depth is estimated by correspondence analysis. To compensate for noisy and not dense depth maps we use a-priori knowledge about the scene to further increase the quality of the reconstruction results. Symbols are assigned to the image content which are used to establish an interpretation of the scene in form of a semantic net. Together with the scene description additional geometric constraints can be selected from a generic knowledge base. Each of these constraints describes a relationship either between parts of the model (e.g. the perpendicularity of two walls) or between the 3-D scene and extracted 2-D image features (e.g. edges or depth information). In the latter case 3-D edges and orientations of model parts are linked by constraints to the respective 2-D feature. Due to noisy data the resulting set of const...

We present a process to improve the structural quality of automatically acquired architectural 3D models. Common architectural features like orientations of walls are exploited. The location of these features is extracted by using a probabilistic technique (RANSAC). The relationships among the features are automatically obtained by labelling them using a semantic net of an architectural scene. An evolutionary algorithm is used to optimise the orientations of the planes. Small irregularities in the planes are removed by projecting the triangulation vertices onto the planes. Planes in the resulting model are aligned to each other. The technique produces models with improved appearance. It is validated on synthetic and real data.

An approach to automatically generate 3--D polygonal models from stereoscopic image pairs of piecewise planar objects is presented. Dense disparity maps are computed by constrained epipolar block matching. Local surface orientations are computed from the quantized disparity map usinga spline approximationunder explicit consideration of quantization noise. The local surface orientation is then clustered into regions of similar surface orientation to find the dominant object planes. A photo realistic 3--D polygonal model of the object is constructed by fitting planar polygons to the surfaces and by mapping original image texture to the model. 1 Introduction The recent advances in multimedia technology and virtual reality applications show that there is a wide range of applications where computer generated 3--D environments are desirable, like in architecture visualization [1], virtual television studios [2], virtual presence for video communications [3] and general "virtual reality" ap...

We present a modular architecture for image understanding and active computer vision which consists of three major components: Sensor and actor interfaces required for data-driven active vision are encapsulated to hide machine-dependent parts; image segmentation is implemented in object-oriented programming as a hierarchy of image operator classes, guaranteeing simple and uniform interfaces; knowledge about the environment is represented either as a semantic network or as statistical object models or as a combination of both; the semantic network formalism is used to represent actions which are needed in explorative vision. We apply

The scene analysis system AIDA is presented. It combines surface reconstruction techniques with object recognition for the generation of 3--D models for computer graphic applications. The system permits an easy way to insert 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 of 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 considering the constraints selected in the prior interpretation. Keywords: Image Processing, Scene Analysis, Knowledge based System, Shape Reconstruction, Object Recognition, Virtual Reality 1 Introduction For the visualization of virtual environments e.g. in applications as flight an...

Applications such as landscape planing, environmental monitoring, and flight and driving simulators have a high demand for realistic landscape models. Quantity, precision and the type of models ask for methods which automate the model generation by evaluation of remote sensing data. The presented modelling system AIDA tackles the demand for efficient representation and high realism by integrating a priori knowledge about the appearance of the objects in the scene to derive object specific constraints for 3D--reconstruction. This requires animage interpretation to assign ameaning to the objects in the scene. For explicit representation of the declarative and procedural knowledge a problem--independent formalism based on semantic nets and rules is used. It provides both a data--driven and model--driven control strategy. 1. Introduction Presently there is a great demand for methods which automate the generation of landscape models from remote sensing data. These models are required for l...

A scene analysis system for automated 3--D modeling of buildings is presented. It combines surface reconstruction techniques with object recognition to generate 3--D models for computer graphics applications. The system permits the insertion of high level constraints, like a specific angle between two house walls, in an explicit knowledge base 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 scene objects using constraints selected in the prior interpretation. 1 Introduction The presented system is designed for the automatical reconstruction of object shapes from digital images. After restoring the 3--D geometry, texture and color of the objects are taken from the original ...