Mario Sormann

Segment-Based Stereo Matching Using Belief Propagation and a Self-Adapting Dissimilarity Measure

A novel stereo matching algorithm is proposed that utilizes color segmentation on the reference image and a self-adapting matching score that maximizes the number of reliable correspondences. The scene structure is modeled by a set of planar surface patches which are estimated using a new technique that is more robust to outliers. Instead of assigning a disparity value to each pixel, a disparity plane is assigned to each segment. The optimal disparity plane labeling is approximated by applying belief propagation. Experimental results using the Middlebury stereo test bed demonstrate the superior performance of the proposed method

Graph Cut Based Multiple View Segmentation for 3D Reconstruction

In this paper we propose a novel framework for efficiently extracting foreground objects in so called short-baseline image sequences. We apply the obtained segmentation to improve subsequent 3D reconstruction results. Essentially, our framework combines a graph cut based optimization algorithm with an intuitive user interface. At first a meanshift segmentation algorithm partitions each image of the sequence into a certain number of regions. Additionally we provide an intelligent graphical user interface for easy specification of foreground as well as background regions across all images of the sequence. Within the graph cut optimization algorithm we define new energy terms to increase the robustness and to keep the segmentation of the foreground object coherent across all images of the sequence. Finally, a refined graph cut segmentation and several adjustment operations allow an accurate and effective foreground extraction. The obtained results are demonstrated on several real world data sets.

Watertight multi-view reconstruction based on volumetric graph-cuts

This paper proposes a fast 3D reconstruction approach for efficiently generating watertight 3D models from multiple short baseline views. Our method is based on the combination of a GPU-based plane-sweep approach, to compute individual dense depth maps and a subsequent robust volumetric depth map integration technique. Basically, the dense depth map values are transformed to a volumetric grid, which are further embedded in a graph structure. The edge weights of the graph are derived from the dense depth map values and if available, from sparse 3D information. The final optimized surface is obtained as a min-cut/max-flow solution of the weighted graph.We demonstrate the robustness and accuracy of our proposed approach on several real world data sets.

High-Performance Multi-View Reconstruction

We present a high performance reconstruction approach, which generates true 3D models from multiple views with known camera parameters. The complete pipeline from depth map generation over depth image integration to the final 3D model visualization is performed on programmable graphics processing units (GPUs). The proposed pipeline is suitable for long image sequences and uses a plane- sweep depth estimation procedure optionally employing robust image similarity functions to generate a set of depth images. The subsequent volumetric fusion step combines these depth maps into an impicit surface representation of the final model, which can be directly displayed using GPU- based ray casting methods. Depending on the number of input views and the desired resolution of the final model the computing times range from several seconds to a few minutes. The quality of the obtained models is illustrated with real-world datasets.

Scanline Optimization for Stereo on Graphics Hardware

In this work we propose a scanline optimization procedure for computational stereo using a linear smoothness cost model performed by programmable graphics hardware. The main idea for an efficient implementation of this dynamic programming approach is a recursive scheme to calculate the min-convolution in a manner suitable for the parallel stream computation model of graphics processing units. Since many image similarity functions can be efficiently calculated by modern graphics hardware, it is reasonable to address the final disparity extraction by graphics processors as well. Our timing results indicate that the proposed approach is beneficial for larger image resolutions and disparity ranges in particular.

Automatic foreground propagation in image sequences for 3D reconstruction

In this paper we introduce a novel method for automatic propagation of foreground objects in image sequences. Our method is based on a combination of the mean-shift operator with the well known intelligent scissors technique. It is effective due to the fact that the images are captured with high overlap, resulting in highly redundant scene information. The algorithm requires an initial segmentation of one image of the sequence as an input. In each consecutive image the segmentation of the previous image is taken as an initialization and the propagation procedure proceeds along four major steps. Each step refines the segmentation of the foreground object and the algorithm converges until all images of the sequence are processed. We demonstrate the effectiveness of our approach on several datasets.

VR modeler: from image sequences to 3D models

In this paper we present a novel interactive modeling system, called <i>VR Modeler,</i> to create 3D geometric models from a set of photographs. In our approach standard automatic reconstruction techniques are assisted by a human operator. The modeling system is efficient and easy to use because the user can concentrate on the 2D segmentation and interpretation of the scene whereas our system is responsible for the corresponding 3D information. Therefore we developed the user interface of VR Modeler as a monocular 3D modeling system. Additionally, we are able to obtain coarse as well as high resolution models from architectural scenes. Finally, we tested the modeling system on different types of datasets to demonstrate the usability of our approach.

Texture mapping for view-dependent rendering

View-dependent multiresolution meshes allow smooth interactive animation and optionally time-critical rendering of huge geometric data-sets and are therefore an important tool for large-model visualization. So far most viewd-ependent rendering frameworks are restricted to models with a topologically simple texture mapping. Our approach overcomes this restriction with a new texturing technique, which allows texture mapping during the runtime simplification process. In fact, novel algorithm generates a graph of textures in a preprocess automatically. This texture graph is furthermore integrated into a view-dependent rendering approach. Particularly we perform a texture validation step for the whole vertex tree, which is the basic data structure in the framework. At runtime the vertex tree is traversed and we introduce a texture proxy map to assure correct texture mapping during view-dependent rendering. Additionally the new technique allows us to guarantee a constant frame rate. Finally the results of our method enhancing visual detail of geometric models are shown.