This paper presents an empirical evaluation of the role of context in a contemporary, challenging object detection task – the PASCAL VOC 2008. Previous experiments with context have mostly been done on home-grown datasets, often with non-standard baselines, making it difficult to isolate the contribution of contextual information. In this work, we present our analysis on a standard dataset, using topperforming local appearance detectors as baseline. We evaluate several different sources of context and ways to utilize it. While we employ many contextual cues that have been used before, we also propose a few novel ones including the use of geographic context and a new approach for using object spatial support. 1.

We describe a novel technique for identifying semantically equivalent parts in images belonging to the same object class, (e.g. eyes, license plates, aircraft wings etc.). The visual appearance of such object parts can differ substantially, and therefore traditional image similarity-based methods are inappropriate for this task. The technique we propose is based on the use of common context. We first retrieve context fragments, which consistently appear together with a given input fragment in a stable geometric relation. We then use the context fragments in new images to infer the most likely position of equivalent parts. Given a set of image examples of objects in a class, the method can automatically learn the part structure of the domain – identify the main parts, and how their appearance changes across objects in the class. Two applications of the proposed algorithm are shown: the detection and identification of object parts and object recognition. 1.

A multi-image focus of attention mechanism has been developed that can quickly distinguish raised objects like buildings from structured background clutter typical to many aerial image scenarios. The underlying approach is the space-sweep stereo method, in which features from multiple images are backprojected onto a virtual, horizontal plane that is methodically swept through the scene. Backprojected gradient orientations from multiple images are highly correlated when they come from scene locations containing structural edges that are roughly horizontal, like building roofs and terrain; otherwise, they tend to be uniformly distributed. These observations are used to define a structural salience measure that can determine whether a given volume of space contains a statistically significant number of structural edges, without first performing precise reconstruction of those edges. The utility of structural salience for computing focus of attention regions is illustrated on sample data f...

Abstract. One of the cornerstones of any intelligent entity is the ability to understand how occurrences in the surrounding world influence its own behaviour. Different states, or situations, in its environment should be taken into account when reasoning or acting. When dealing with different situations, context is the key element used to infer possible actions and information needs. The activities of the perceiving agent and other entities are arguably one of the most important features of a situation; this is equally true whether the agent is artificial or not. This work proposes the use of Activity Theory to first model context and further on populate the model for assessing situations in a pervasive computing environment. Through the socio-technical perspective given by Activity Theory, the knowledge intensive context model, utilised in our ambient intelligent system, is designed. 1

Image understanding (IU) techniques for automatic site reconstruction have demonstrated success within restricted domains and for small numbers of model classes. However, these techniques often fail when applied out of context and do not "scale-up" into a more general solution. Under the APGD program, we are constructing a knowledgebased site reconstruction system that automatically selects the correct algorithm according to the current context, applies it to a focused subset of the data, and constrains the interpretation of the result through the explicit use of knowledge. 1 Introduction The extraction and reconstruction of building models from aerial images has become an important area of research in recent years. Significant progress has been made and several systems perform reasonably well within their appropriate domains [Collins'95, Herman'94, Lin et al.'94, Chellapa et al.'94]. For example, recent testing of the Ascender I system has shown it capable of automatically extractin...

The Automated Site Construction, Extension, Detection and Refinement system (ASCENDER) has been developed to automatically populate a site model with buildings extracted from multiple, overlapping views. Version 1.0 of the system has been delivered for evaluation on classified imagery. Evaluation results on an unclassified Ft.Hood data set are presented here. Extensions to the system that allow it to detect a wide range of building classes, including peaked roof and multi-level flat roofed structures are described. Recent work on symbolic extraction of surface structures such as windows greatly enhances the visual realism of graphical site model displays. 1 Introduction The Research and Development for Image Understanding Systems (RADIUS) project is a national effort to apply image understanding (IU) technology to support model-based aerial image analysis [5]. Automated construction and management of 3D geometric site models enables efficient exploitation of the tremendous volume of ...

This paper presents interim results from an ongoing project on aerial image reconstruction. One important task in image interpretation is the process of understanding and identifying segments of an image. In this effort a knowledge based vision system is being presented, where the selection of IU algorithms and the fusion of information provided by them is combined in an efficient way. Knowledge based vision systems developed so far have focused on the interpretation problem for a small set of object classes. A major problem with these systems is that the knowledge base, control mechanism and knowledge sources are combined into a single intertwined system and the addition of new knowledge or change of domain requires a significant efforts. In our current work, the knowledge base and control mechanism (reasoning subsystem) are independent of the knowledge sources (visual subsystem). This gives the system the flexibility to add or change knowledge sources with only minor changes in the r...

This paper deals with the integration of image understanding (IU) programs using a knowledgebased approach. The basic concepts of program integration are discussed, and a simple problem-solving model for program integration is outlined. Two types of reasoning, planning and execution control, are identified. A system developed using this model, called OCAPI (Optimizing, Controlling and Automating the Processing of Images), is introduced. OCAPI is in an AI environment in which the reasoning used by the IU specialist is formally represented using frames and production rules. An example of an application developed using OCAPI is presented, and the advantages and shortcomings of this approach are discussed. The support of the Advanced Research Projects Agency (ARPA Order No. ????) is gratefully acknowledged, as is the help of Sandy German in preparing this paper. 1 Introduction In any rapidly-evolving field of research such as image understanding (IU), the development of new methods is of...

We propose a methodology for the development of image understanding systems that provide both convenience and flexibility. In this methodology, the image analyst provides the input data, specifies the IU task to be performed, and then provides feedback in the form of qualitative evaluations of the result(s) obtained. These assessments are interpreted in a knowledgebased framework to select the best algorithms and to find the most suitable parameter settings. In this manner the IU system is given the capacity to tune itself for optimal performance. A sample application (vehicle detection in aerial imagery) is developed to illustrate the approach. 1 Introduction Image Understanding (IU) systems used in challenging operational environments should satisfy the conflicting requirements of flexibility and convenience. Flexibility is the ability of the system to accommodate variations in the characteristics of the input data. Convenience means that the system can be operated by an image analy...

Over the last several years the concept of modelsupported exploitation(MSE) has evolved to a point where relatively simple computer vision algorithms can extract significant intelligence information from aerial images in a robust and reliable manner. Information extraction is enabled by the use of detailed 3D site models which provide an extensive context for the application of image analysis algorithms. This paper reviews the basic MSE concept and illustrates the approach using three operational concepts taken from the RADIUS project, quick-look, detection and counting and focussed change detection. 1 Model supported exploitation 1.1 The site model A project to support increased productivity for image analysts called RADIUS(Research and Development for Image Understanding Systems) has been under development over the last several years[1]. The central concept of RADIUS is model supported exploitation (MSE). In this concept a 3D model is constructed from a number of images of a site. ...