Abstract—Computing 3D object descriptions from images is an important goal of computer vision. A key problem here is the evaluation of a hypothesis based on evidence that is uncertain. There have been few efforts on applying formal reasoning methods to this problem. In multiview and multimode object description problems, reasoning is required on evidence features extracted from multiple images and nonintensity data. One challenge here is that the number of the evidence features varies at runtime because the number of images being used is not fixed and some modalities may not always be available. We introduce an augmented Bayesian network, the expandable Bayesian network (EBN), which instantiates its structure at runtime according to the structure of input. We introduce the use of hidden variables to handle correlation of evidence features across images. We show an application of an EBN to a multiview building description system. Experimental results show that the proposed method gives significant and consistent performance improvement to others. Index Terms—Multiview object description, learning, uncertain reasoning, building description, Bayesian network. 1

Producing data of known quality is an essential operation of mapping agencies such as Ordnance Survey. For these data to be of value to our customers, we need to understand what quality measures will allow them to assess whether the data are fit for their purpose. In the case of 3-dimensional (3D) data, this is particularly important as it will inform research into capturing and modelling these data. However, for a data type in its infancy, such as 3D data, it is rare that a clear idea of quality requirements is available since the full range of uses of the data is still unknown. Instead, the potential use contexts of such data need to be investigated. To this end, we have conducted user needs research across a wide range of professional use contexts. This research has been analysed to identify measures and their required quality for use contexts where 3D information about buildings is of particular interest to the user. However, it is often the case that the user cannot realistically make explicit statements anticipating what they would require in terms of 3D data measures and quality elements such as positional accuracy. Instead, it is possible to identify 3D building data characteristics and quality tolerances from implicit statements about use context and objectives from interviews with a wide range of professionals. Characteristics identified include the highest point of a structure and the maximum height of roof ridge, and others such as the the geometric shapes of roofs, buildings and the space between them, which will clearly present some challenges for developing usable quality measures. Preliminary results of this research are presented. 1

ABSTRACT: We present an approach to detecting and describing compositions of buildings with flat or complex rooftops by using multiple, overlapping images of the scene. First, 3-D features are generated by using multiple images, and rooftop hypotheses are generated by neighborhood searches on those features. For robust generation of 3-D features, we present a probabilistic approach to address the epipolar alignment problem in line matching. Image-derived unedited elevation data is used to assist feature matching, and to generate rough cues of the presence of 3-D structures. These cues help reduce the search space significantly. Experimental results are shown on some complex buildings. 1

In the real world, systems/processes often evolve without fixed and predictable dynamic models. To represent such applications we need uncertainty models, like Bayesian Nets (BN) that are formed online and in a self-evolving data-driven way. But current BN frameworks cannot handle simultaneous scalability in the model structure and causal relations. We show how current BNs fail in different applications from several fields, ranging from computer vision to database retrieval to medical diagnostics. We propose a novel Structure Modifiable Adaptive Reason-building Temporal Bayesian Networks (SmartBN) that has scalability for uncertainty in both, structures and causal relations. We evaluate its performance for a 3D model building application for vehicles in traffic video. 1.

High quality 3D building databases are essential inputs for urban area Geographic Information Systems. Since manual generation of these databases is very costly and time consuming, the development of automated algorithms is of great need. This article presents a new algorithm to automatically extract accurate and reliable 3D building information. High overlapping aerial images are used as input to the algorithm. Radiometric and geometric properties of buildings are utilized to distinguish building roofs in the images. This is accomplished using image segmentation and neural network techniques. A rule-based system is used to extract the vertices of the roof polygons in all images. The 3D coordinates of these vertices are computed using photogrammetric mathematical models. The algorithm is tested on a number of buildings in a complex urban scene. Results showed a detection rate of 99 % and a false alarm rate of 5.0%. The root mean square error for the extracted building vertices is 0.25 meter using 1:4000 scale aerial photographs scanned at 30 micron.