Panoramas create summary views of multiple images, which make them a valuable means of analyzing huge quantities of image and video data. This paper introduces the Ray Graph - a general framework for panorama construc...
详细信息
Panoramas create summary views of multiple images, which make them a valuable means of analyzing huge quantities of image and video data. This paper introduces the Ray Graph - a general framework for panorama construction. With rays as its vertices, the Ray Graph uses its edges to specify a set of coherency relationships among all input rays. Consequently, by using a set of simple graph traversal rules, a diverse set of panorama structures can be enumerated, which can be used to efficiently and robustly generate panoramic images from image collections. To demonstrate this framework, we first use it to recreate both 360 degree and street panoramas. We further introduce two new panorama models, the centipede panorama - a hybrid of 360 degree and street panoramas, and the storytelling panorama - a time encoding panorama. Finally, we demonstrate the flexibility of this framework by enabling interactive brushing of panoramic regions for removal of undesired features such as occlusions and moving objects.
With increasing speed of graphics rendering, the bottleneck in walkthrough applications has shifted to data transfer from secondary storage device to main memory. While techniques designed to reduce the data transfer ...
详细信息
With increasing speed of graphics rendering, the bottleneck in walkthrough applications has shifted to data transfer from secondary storage device to main memory. While techniques designed to reduce the data transfer volume and amortize the transfer cost are well-studied, the disk seek time, which is one of the most important components of the total rendering cost is not reduced explicitly. In this work, we propose an orthogonal approach to address the disk seek time bottleneck, namely single-seek data layouts. This is a solution in one end of the spectrum of solutions that guarantee an upper bound on the number of disk seeks. Using this approach, we can reduce the number of disk seeks required to load the data for any viewpoint in the scene to no more than one. We achieve this single seek layout using data redundancy. We provide a theoretical proof on the upper-bound of this redundancy factor, and analyze its trade-off with the rendering performance through an implementation that uses this data layout for walkthrough applications of datasets with hundreds of millions of triangles.
The proceedings contain 38 papers. The topics discussed include: automatic produce classification from images using color, texture and appearance cues;face recognition aiding historical photographs indexing using a tw...
ISBN:
(纸本)9780769533582
The proceedings contain 38 papers. The topics discussed include: automatic produce classification from images using color, texture and appearance cues;face recognition aiding historical photographs indexing using a two-stage training scheme and an enhanced distance measure;a new training algorithm for pattern recognition technique based on straight line segments;crop type recognition based on Hidden Markov Models of plant phenology;synchronizing video cameras with non-overlapping fields of view;people detection under occlusion in multiple camera views;occlusion handling for object tracking using a fast level set method;efficient text color modulation for printed side communications and data hiding;estimating the skew angle of scanned document through background area information;and benchmark for quantitative evaluation of assisted object segmentation methods to image sequences.
In the framework of the image Foresting Transform (IFT), there is a class of connectivity functions that were vaguely explored, which corresponds to the non-smooth connectivity functions (NSCF). These functions are mo...
详细信息
In the framework of the image Foresting Transform (IFT), there is a class of connectivity functions that were vaguely explored, which corresponds to the non-smooth connectivity functions (NSCF). These functions are more adaptive to cope with the problems of field in homogeneity, which are common in MR images of 3 Tesla. In this work, we investigate the NSCF from the standpoint of theoretical and experimental aspects. We formally classify several non-smooth functions according to a proposed diagram representation. Then, we investigate some theoretical properties for some specific regions of the diagram. Our analysis reveals that many NSCFs are, in fact, the result of a sequence of optimizations, each of them involving a maximal set of elements, in a well-structured way. Our experimental results indicate that substantial improvements can be obtained by NSCFs in the 3D segmentation of MR images of 3 Tesla, when compared to smooth connectivity functions.
View-dependent image-based rendering techniques have become increasingly popular as they combine the high quality of images with the explorability of interactive techniques. However, in the context of volume rendering...
详细信息
View-dependent image-based rendering techniques have become increasingly popular as they combine the high quality of images with the explorability of interactive techniques. However, in the context of volume rendering, previous approaches suffer from various shortcomings, including the limitation to surfaces, expensive generation, and insufficient occlusion and motion parallax impairing depth perception. In this paper, we propose Volumetric Depth images (VDI) to overcome these issues for view-dependent volume visualization by an extension of the Layered Depth image (LDI) approach. Instead of only saving for each view ray of one camera configuration the depth and color values for a set of surfaces, as in LDIs, VDIs store so-called super segments, each consisting of a depth range as well as composited color and opacity. This compact representation is independent from the structure of the original data and can be generated by slight modification of ray casters with very low overhead. VDIs can be rendered efficiently at high quality with arbitrary camera configurations by means of proxy frustum geometry and an efficient depth ordering scheme. When viewing the scene from the initial view point, VDIs produce results identical to the original ray casting. As demonstrated by means of a prototype implementation and data from different fields, our approach can be useful for preview rendering and a-priori analysis in in-situ contexts among others.
Ray tracing allows the rendering of scenes with very complex light interactions. It is based on the idea that reflection, refraction and shadows can be modeled by recursively following the path that light takes as it ...
详细信息
Ray tracing allows the rendering of scenes with very complex light interactions. It is based on the idea that reflection, refraction and shadows can be modeled by recursively following the path that light takes as it bounces through an environment. However, despite its conceptual simplicity, tracing rays is a computationally intensive task. Also, optimizing memory management to increase efficiency is hard since coherent access in 3D space would not generate coalescent memory patterns. We present a new caching-like strategy suitable for real-time ray tracing which is capable to store data generated in previous frames in such a way that coherent memory access is achieved while data is reused by subsequent frames. By storing light bounce results of previously traced rays in a cube map attached to each scene object, we show that it is possible to explore the efficient memory sampling mechanism provided by the graphics hardware to increase frame rate. Our approach is suitable for static scenes and may prevent deep interactions of rays with the scene as well as enable synchronous computation of rays in parallelized architectures, and it can be easily integrated to any existing ray tracing solutions.
In this paper we propose a visual estimation algorithm to aid the guidance system of an unmanned aerial vehicle (UAV). Specifically, we are considering a glider UAV trying to reach a target zone. Given that the UAV is...
详细信息
In this paper we propose a visual estimation algorithm to aid the guidance system of an unmanned aerial vehicle (UAV). Specifically, we are considering a glider UAV trying to reach a target zone. Given that the UAV is falling fast, the scale of the images captured by the on-board camera grows. The vehicle dynamics and the control algorithm cause fast changes in the region seen by the camera and in the way it projects to the image plane. We propose an algorithm based on projective transformations to estimate the target coordinate during the flight trajectory. The proposed vision algorithm provides accurate position information allowing the control system to guide the UAV to the target zone. Simulations have shown the accuracy of the vision algorithm and its usefulness in the considered application.
In recent years, many approaches for real-time simulation of physical phenomena using particles have been proposed. Many of these use 3D grids for representing spatial distributions and employ a collision detection te...
详细信息
In recent years, many approaches for real-time simulation of physical phenomena using particles have been proposed. Many of these use 3D grids for representing spatial distributions and employ a collision detection technique where particles must be sorted with respect to the cells they occupy. In this paper we propose several techniques that make it possible to explore spatio-temporal coherence in order to reduce the work needed to produce a correct ordering and thus accelerate the collision detection phase of the simulation. Sequential and GPU-based implementations are discussed, and experimental results are presented. Although devised with particle-based simulations in mind, the proposed techniques have a broader scope, requiring only some means of establishing subsequences of the input which did not change from one frame to the next.
This work addresses the problem of aligning a set of point clouds acquired by an RGB-D sensor. To achieve this, we propose the fusion of RGB-D data with the orientation provided by a MARG (Magnetic, Angular Rate and G...
详细信息
This work addresses the problem of aligning a set of point clouds acquired by an RGB-D sensor. To achieve this, we propose the fusion of RGB-D data with the orientation provided by a MARG (Magnetic, Angular Rate and Gravity) sensor, a combination that hasn't been extensively explored in the literature. Our methodology uses MARG data both in the coarse pair wise alignment between point clouds and in the loop closure detection between key frames. In our experiments, we were able to align the walls of a room with dimensions 9.84m x 7.13m. Our analysis shows us that MARG data helps to improve the alignment quality.
We think that state-of-the-art techniques in computergraphics and geometry processing can be leveraged in training and entertainment to make the topic of cultural heritage more accessible to a wider audience. In a co...
详细信息
We think that state-of-the-art techniques in computergraphics and geometry processing can be leveraged in training and entertainment to make the topic of cultural heritage more accessible to a wider audience. In a cooperation with the “Antikensammlung” in Erlangen we produced five unique applications - all based around emperor Augustus - to visualize different scientific aspects for a big event targeted towards the general public. The applied methods include blending, geometric fitting, animation transfer and visualization techniques. Besides being entertaining, some of the presented applications are the foundation for more substantial research. (For our results video please visit http://***/uploads/video/***).
暂无评论