No description available.

In this thesis, we propose a new representation, which we name Interaction Bisector Surface (IBS), that can describe the general nature of spatial relationship. We show that the IBS can be applied in 3D scene analysis, retrieval and synthesis. Despite the fact that the spatial relationship between different objects plays a significant role in describing the context, few works have focused on elaborating a representation that can describe arbitrary interactions between different objects. Previous methods simply concatenate the individual state vectors to produce a joint space, or only use simple representations such as relative vectors or contacts to describe the context. Such representations do not contain detailed information of spatial relationships. They cannot describe complex interactions such as hooking and enclosure. The IBS is a data structure with rich information about the interaction. It provides the topological, geometric and correspondence features that can be used to classify and recognize interactions. The topological features are at the most abstract level and it can be used to recognize spatial relationships such as enclosure, hooking and surrounding. The geometric features encode the fine details of interactions. The correspondence feature describes which parts of the scene elements contribute to the interaction and is especially useful for recognizing character-object interactions. We show examples of successful classification and retrieval of different types of data including indoor static scenes and dynamic scenes which contain character-object interactions. We also conduct an exhaustive comparison which shows that our method outperforms existing approaches. We also propose a novel approach to automatically synthesizing new interactions from example scenes and new objects. Given an example scene composed of two objects, the open space between the objects is abstracted by the IBS. Then, an translation, rotation and scale equivariant feature called shape coverage feature, which encodes how the point in the open space is surrounded by the environment, is computed near the IBS and around the open space of the new objects. Finally, a novel scene is synthesized by conducting a partial matching of the open space around the new objects with the IBS. Using our approach, new scenes can be automatically synthesized from example scenes and new objects without relying on label information, which is especially useful when the data of scenes and objects come from multiple sources.

Perception is a key feature in how any creature or autonomous system relates to its environment. While there are many types of perception, this thesis focuses on the improvement of the visual robotics perception systems. By implementing a broadband passive sensing system in conjunction with current perception algorithms, this thesis explores scene reconstruction and world modeling. The process involves two main steps. The first is stereo correspondence using block matching algorithms with filtering to improve the quality of this matching process. The disparity maps are then transformed into 3D point clouds. These point clouds are filtered again before the registration process is done. The registration uses a SAC-IA matching technique to align the point clouds with minimum error.  The registered final cloud is then filtered again to smooth and down sample the large amount of data. This process was implemented through software architecture that utilizes Qt, OpenCV, and Point Cloud Library. It was tested using a variety of experiments on each of the components of the process.  It shows promise for being able to replace or augment existing UGV perception systems in the future. / Master of Science

This dissertation addresses a number of inter-related and fundamental problems in computer vision. Specifically, we address object discovery, recognition, segmentation, and 3D pose estimation in images, as well as 3D scene reconstruction and scene interpretation. The key ideas behind our approaches include using shape as a basic object feature, and using structured prediction modeling paradigms for representing objects and scenes. In this work, we make a number of new contributions both in computer vision and machine learning. We address the vision problems of shape matching, shape-based mining of objects in arbitrary image collections, context-aware object recognition, monocular estimation of 3D object poses, and monocular 3D scene reconstruction using shape from texture. Our work on shape-based object discovery is the first to show that meaningful objects can be extracted from a collection of arbitrary images, without any human supervision, by shape matching. We also show that a spatial repetition of objects in images (e.g., windows on a building facade, or cars lined up along a street) can be used for 3D scene reconstruction from a single image. The aforementioned topics have never been addressed in the literature. The dissertation also presents new algorithms and object representations for the aforementioned vision problems. We fuse two traditionally different modeling paradigms Conditional Random Fields (CRF) and Random Forests (RF) into a unified framework, referred to as (RF)^2. We also derive theoretical error bounds of estimating distribution ratios by a two-class RF, which is then used to derive the theoretical performance bounds of a two-class (RF)^2. Thorough experimental evaluation of individual aspects of all our approaches is presented. In general, the experiments demonstrate that we outperform the state of the art on the benchmark datasets, without increasing complexity and supervision in training. / Graduation date: 2011 / Access restricted to the OSU Community at author's request from May 12, 2011 - May 12, 2012

In this thesis, the extraction of sparse 3D points from calibrated image sequences is studied. The presented method for sparse 3D reconstruction is examined in two steps, where the first part addresses the problem of two-view reconstruction, and the second part is the extension of the two-view reconstruction algorithm to include multiple views. The examined two-view reconstruction method consists of some basic building blocks, such as feature detection and matching, epipolar geometry estimation, and the reconstruction of cameras and scene structure. Feature detection and matching is achieved by Scale Invariant Feature Transform (SIFT) method. For the estimation of epipolar geometry, the 7-point and 8-point algorithms are examined for Fundamental matrix (F-matrix) computation, while RANSAC and PROSAC are utilized for the robustness and accuracy for model estimation. In the final stage of two-view reconstruction, the camera projection matrices are computed from the F-matrix, and the locations of 3D scene points are estimated by triangulation / hence, determining the scene structure and cameras up to a projective transformation. The extension of the two-view reconstruction to multiple views is achieved by estimating the camera projection matrix of each additional view from the already reconstructed matches, and then adding new points to the scene structure by triangulating the unreconstructed matches. Finally, the reconstruction is upgraded from projective to metric by a rectifying homography computed from the camera calibration information. In order to obtain a refined reconstruction, two different methods are suggested for the removal of erroneous points from the scene structure. In addition to the examination of the solution to the reconstruction problem, experiments have been conducted that compare the performances of competing algorithms used in various stages of reconstruction. In connection with sparse reconstruction, a rate-distortion efficient piecewise planar scene representation algorithm that generates mesh models of scenes from reconstructed point clouds is examined, and its performance is evaluated through experiments.

In this thesis, the structure from motion problem for calibrated scenes containing independently moving objects (IMO) has been studied. For this purpose, the overall reconstruction process is partitioned into various stages. The first stage deals with the fundamental problem of estimating structure and motion by using only two views. This process starts with finding some salient features using a sub-pixel version of the Harris corner detector. The features are matched by the help of a similarity and neighborhood-based matcher. In order to reject the outliers and estimate the fundamental matrix of the two images, a robust estimation is performed via RANSAC and normalized 8-point algorithms. Two-view reconstruction is finalized by decomposing the fundamental matrix and estimating the 3D-point locations as a result of triangulation. The second stage of the reconstruction is the generalization of the two-view algorithm for the N-view case. This goal is accomplished by first reconstructing an initial framework from the first stage and then relating the additional views by finding correspondences between the new view and already reconstructed views. In this way, 3D-2D projection pairs are determined and the projection matrix of this new view is estimated by using a robust procedure. The final section deals with scenes containing IMOs. In order to reject the correspondences due to moving objects, parallax-based rigidity constraint is used. In utilizing this constraint, an automatic background pixel selection algorithm is developed and an IMO rejection algorithm is also proposed. The results of the proposed algorithm are compared against that of a robust outlier rejection algorithm and found to be quite promising in terms of execution time vs. reconstruction quality.

We address the problem of estimating a high quality dense depth map from a single RGB input image. We start out with a baseline encoder-decoder convolutional neural network architecture and pose the question of how the global processing of information can help improve overall depth estimation. To this end, we propose a transformer-based architecture block that divides the depth range into bins whose center value is estimated adaptively per image. The final depth values are estimated as linear combinations of the bin centers. We call our new building block AdaBins. Our results show a decisive improvement over the state-of-the-art on several popular depth datasets across all metrics. We also validate the effectiveness of the proposed block with an ablation study.

Optical three dimensional (3D) mapping routines are used in inspection robots to detect faults by creating 3D reconstructions of environments. To detect surface faults, sub millimeter depth resolution is required to determine minute differences caused by coating loss and pitting. Sensors that can detect these small depth differences cannot quickly create contextual maps of large environments. To solve the 3D mapping problem, a sensor fused approach is proposed that can gather contextual information about large environments with one depth sensor and a SLAM routine; while local surface defects can be measured with an actuated optical profilometer. The depth sensor uses a modified Kinect Fusion to create a contextual map of the environment. A custom actuated optical profilometer is created and then calibrated. The two systems are then registered to each other to place local surface scans from the profilometer into a scene context created by Kinect Fusion. The resulting system can create a contextual map of large scale features (0.4 m) with less than 10% error while the optical profilometer can create surface reconstructions with sub millimeter resolution. The combination of the two allows for the detection and quantification of surface faults with the profilometer placed in a contextual reconstruction. / Master of Science

This thesis work deals with the most actual topic in Computer Vision field which is scene understanding and this using matching of 3D feature point images. The objective is to make use of Saab’s latest breakthrough in extraction of 3D feature points, to identify the best alignment of at least two 3D feature point images. The thesis gives a theoretical overview of the latest algorithms used for feature detection, description and matching. The work continues with a brief description of the simultaneous localization and mapping (SLAM) technique, ending with a case study on evaluation of the newly developed software solution for SLAM, called slam6d. Slam6d is a tool that registers point clouds into a common coordinate system. It does an automatic high-accurate registration of the laser scans. In the case study the use of slam6d is extended in registering 3D feature point images extracted from a stereo camera and the results of registration are analyzed. In the case study we start with registration of one single 3D feature point image captured from stationary image sensor continuing with registration of multiple images following a trail. Finally the conclusion from the case study results is that slam6d can register non-laser scan extracted feature point images with high-accuracy in case of single image but it introduces some overlapping results in the case of multiple images following a trail.

Goal of this work is to write an optimized program for visualization of 3D scenes using ray tracing method. First, the theory of ray tracing together with particular techniques are presented. Next part focuses on different approaches to accelerate the algorithm. These are space partitioning structures, fast ray-triangle intersection technique and possibilities to parallelize the whole ray tracing method. A standalone chapter addresses the design and implementation of the ray tracing program.