Claus-E. Liedtke

Hierarchical 3D Pose Estimation for Articulated Human Body Models from a Sequence of Volume Data

This contribution describes a camera-based approach to fully automatically extract the 3D motion parameters of persons using a model based strategy. In a first step a 3D body model of the person to be tracked is constructed automatically using a calibrated setup of sixteen digital cameras and a monochromatic background. From the silhouette images the 3D shape of the person is determined using the shape-from-silhouette approach. This model is segmented into rigid body parts and a dynamic skeleton structure is fit. In the second step the resulting movable, personalized body template is exploited to estimate the 3D motion parameters of the person in arbitrary poses. Using the same camera setup and the shape-from-silhouette approach a sequence of volume data is captured to which the movable body template is fit. Using a modified ICP algorithm the fitting is performed in a hierarchical manner along the the kinematic chains of the body model. The resulting sequence of motion parameters for the articulated body model can be used for gesture recognition, control of virtual characters or robot manipulators.

Segmentation of blood smears by hierarchical thresholding

Abstract A major problem in the development of systems for automated differential blood count of leucocytes in the peripheral blood is the correct segmentation of the cell scene, i.e., its decomposition into nucleus, plasma, erythrocytes, thrombocytes, etc. Several algorithms have been proposed in the literature and they differ considerably in performance, expense of implementation, and speed. This paper describes a fast and simple segmentation scheme based on hierarchical thresholding. The thresholds are derived from histograms of the cell scene based on two color features used at different processing steps. Segmentation is achieved by application of several local operators to the cell scene. These operators can easily be implemented in hardware. Therefore the segmentation method is suitable for the implementation on a special fast hardware processor as part of a system for automated differential blood counting.

Architecture of the knowledge based configuration system for image analysis 'CONNY'

For successful image analysis it is necessary that the image analysis system be configured to meet the requirements of the specific task and the specific data material. This includes the selection of the operators and the adaptation of the free parameters. The system CONNY has been developed which performs this configuration process automatically based on a user-specified task definition and the knowledge of an image analysis expert. The experts knowledge has been assessed, stored and used by employing different paradigms of explicit knowledge representation.<<ETX>>

Use of Explicit Knowledge for the Reconstruction of 3-D Object Geometry

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.

A framework for GIS and imagery data fusion in support of cartographic updating

Abstract The task to update map databases regularly using images calls for automation. Within this process GIS and image data have to be combined. The different nature and content of these information sources prevent a direct comparison. In this paper an approach for the combination of GIS data and aerial images is proposed. We make use of a knowledge base, modelling the objects which are expected to be found. This leads to a refined image interpretation process and enables a revision of the GIS data. The focus is thereby on settlement and industrial areas which are detected automatically from images. The use of a semantic network allows the formulation of such complex objects expected in the image and supports decisions on a high level of abstraction. Reasoning is supported by using the existing GIS information for hypothesis generation. The decisions combine different clues which evolve during the image analysis process. The analysis generates a hierarchical scene description for the area of study and evaluates the correctness of the GIS data.

A Knowledge-Based System for Context Dependent Evaluation of Remote Sensing Data

Automatic interpretation of remote sensing data gathers more and more importance for surveillance tasks, reconnaissance and automatic generation and quality control of geographic maps. Methods and applications exist for structural analysis of image data as well as specialized segmentation algorithms for certain object classes. At the Institute of Communication Theory and Signal Processing focus is set on procedures that incorporate a priori knowledge into the interpretation process. Though many advanced image processing algorithms have been developed in the past, a disadvantage of earlier interpretation systems is the missing combination capability for the results of different - especially multisensor - image processing operators. The system GEOAIDA presented in this paper utilizes a semantic net to model a priori knowledge about the scene. The low-level, context dependent segmentation is accomplished by already existing, external image processing operators, which are integrated and controlled by GEOAIDA. Also the evaluation of the interpretation hypothesis is done by externalop erators, linked to the GEOAIDA system. As a result an interactive map with user selectable level-of-detail is generated.

A region extraction approach to blood smear segmentation

Abstract A major problem in the development of systems for automated differential blood count of leucocytes of the peripheral blood is the correct segmentation of the cell scene. Standard preparation techniques of blood smears have been optimized for the visual inspection of the cell scene. Variations in the preparation have a negligible effect on the scene evaluation by visual analysis but do pose major problems on an automated analysis. Therefore a new algorithm for the segmentation of microscopic cell scenes of leucocytes has been developed which uses, in addition to local properties of the scene, global information about neighborhood relations and shape of the scene components for the improvement of the segmentation. The algorithm is based on a region extraction approach and a subsequent labeling procedure. A priori knowledge about shape and neighborhood relations is entered by a relaxation process.

Automated modelling of real human faces for 3D animation

This paper describes the automated creation of a textured face mask of a natural person suitable for real time animation. A 3D surface polygon face mask is used. A set of muscles is defined to animate the mask. The mask is adapted to a real person using 3D surface data from a calibrated stereo sensor. First, the facial features of the mask are aligned to the 3D data by scaling the mask. Next, a local shape adaptation moves the mask vertices along their surface normals to be coplanar with the measured 3D surface. The adapted mask is textured from one of the stereo images yielding a highly realistic impression. The predefined muscles are used to animate the textured mask.

Shape adaptation for modelling of 3D objects in natural scenes

Rigid 3D objects were modelled automatically from an image sequence taken by a camera that was rotated around the object. The image sequence was recorded using a calibrated camera which allows one to measure the camera positions and to estimate the true object size. The 3D object shape was obtained in two steps. The object silhouettes were employed to find the enclosing volume of the object. The volume was converted into a flexible surface representation and the 3D shape was refined based on the texture information of the object surface. Texture mapping was applied to generate a highly realistic 3D model of the object.<<ETX>>

3D Reconstruction of Volume Defects from Few X-Ray Images

In nondestructive testing for quality control of industrial objects the standard X-ray analysis produces a 2D projection of the 3D objects. Defects can be detected but cannot be localized in 3D position, size and shape. Tomographic testing equipment turns frequently out to be too costly and time consuming for many applications. Here a new approach for 3D reconstruction is suggested using standard X-ray equipment without costly positioning equipment. The new approach requires only a small number of X-ray views from different directions in order to reduce the image acquisition time. The geometric and photometric imaging properties of the system are calibrated using different calibration patterns. The parameters of a CAHV camera model are obtained for each view permitting the exact registration of the acquired images. The effciency of the 3D reconstruction algorithm has been increased by limiting the reconstruction to regions of interest around the defects. This requires an automated segmentation. The 3D reconstruction of the defects is performed with an iterative procedure. Regularization of the reconstruction problem is achieved on the basis of the maximum entropy principle. The reliability and robustness of the method has been tested on simulated and real data.