Chavdar Papazov

An efficient RANSAC for 3D object recognition in noisy and occluded scenes

In this paper, we present an efficient algorithm for 3D object recognition in presence of clutter and occlusions in noisy, sparse and unsegmented range data. The method uses a robust geometric descriptor, a hashing technique and an efficient RANSAC-like sampling strategy. We assume that each object is represented by a model consisting of a set of points with corresponding surface normals. Our method recognizes multiple model instances and estimates their position and orientation in the scene. The algorithm scales well with the number of models and its main procedure runs in linear time in the number of scene points. Moreover, the approach is conceptually simple and easy to implement. Tests on a variety of real data sets show that the proposed method performs well on noisy and cluttered scenes in which only small parts of the objects are visible.

Rigid 3D geometry matching for grasping of known objects in cluttered scenes

In this paper, we present an efficient 3D object recognition and pose estimation approach for grasping procedures in cluttered and occluded environments. In contrast to common appearance-based approaches, we rely solely on 3D geometry information. Our method is based on a robust geometric descriptor, a hashing technique and an efficient, localized RANSAC-like sampling strategy. We assume that each object is represented by a model consisting of a set of points with corresponding surface normals. Our method simultaneously recognizes multiple model instances and estimates their pose in the scene. A variety of tests shows that the proposed method performs well on noisy, cluttered and unsegmented range scans in which only small parts of the objects are visible. The main procedure of the algorithm has a linear time complexity resulting in a high recognition speed which allows a direct integration of the method into a continuous manipulation task. The experimental validation with a seven-degree-of-freedom Cartesian impedance controlled robot shows how the method can be used for grasping objects from a complex random stack. This application demonstrates how the integration of computer vision and soft-robotics leads to a robotic system capable of acting in unstructured and occluded environments.

Stochastic global optimization for robust point set registration

In this paper, we propose a new algorithm for pairwise rigid point set registration with unknown point correspondences. The main properties of our method are noise robustness, outlier resistance and global optimal alignment. The problem of registering two point clouds is converted to a minimization of a nonlinear cost function. We propose a new cost function based on an inverse distance kernel that significantly reduces the impact of noise and outliers. In order to achieve a global optimal registration without the need of any initial alignment, we develop a new stochastic approach for global minimization. It is an adaptive sampling method which uses a generalized BSP tree and allows for minimizing nonlinear scalar fields over complex shaped search spaces like, e.g., the space of rotations. We introduce a new technique for a hierarchical decomposition of the rotation space in disjoint equally sized parts called spherical boxes. Furthermore, a procedure for uniform point sampling from spherical boxes is presented. Tests on a variety of point sets show that the proposed registration method performs very well on noisy, outlier corrupted and incomplete data. For comparison, we report how two state-of-the-art registration algorithms perform on the same data sets.

Deformable 3D Shape Registration Based on Local Similarity Transforms

In this paper, a new method for deformable 3D shape registration is proposed. The algorithm computes shape transitions based on local similarity transforms which allows to model not only as‐rigid‐as‐possible deformations but also local and global scale. We formulate an ordinary differential equation (ODE) which describes the transition of a source shape towards a target shape. We assume that both shapes are roughly pre‐aligned (e.g., frames of a motion sequence). The ODE consists of two terms. The first one causes the deformation by pulling the source shape points towards corresponding points on the target shape. Initial correspondences are estimated by closest‐point search and then refined by an efficient smoothing scheme. The second term regularizes the deformation by drawing the points towards locally defined rest positions. These are given by the optimal similarity transform which matches the initial (undeformed) neighborhood of a source point to its current (deformed) neighborhood. The proposed ODE allows for a very efficient explicit numerical integration. This avoids the repeated solution of large linear systems usually done when solving the registration problem within general‐purpose non‐linear optimization frameworks. We experimentally validate the proposed method on a variety of real data and perform a comparison with several state‐of‐the‐art approaches.

Predicting human intention in visual observations of hand/object interactions

The main contribution of this paper is a probabilistic method for predicting human manipulation intention from image sequences of human-object interaction. Predicting intention amounts to inferring the imminent manipulation task when human hand is observed to have stably grasped the object. Inference is performed by means of a probabilistic graphical model that encodes object grasping tasks over the 3D state of the observed scene. The 3D state is extracted from RGB-D image sequences by a novel vision-based, markerless hand-object 3D tracking framework. To deal with the high-dimensional state-space and mixed data types (discrete and continuous) involved in grasping tasks, we introduce a generative vector quantization method using mixture models and self-organizing maps. This yields a compact model for encoding of grasping actions, able of handling uncertain and partial sensory data. Experimentation showed that the model trained on simulated data can provide a potent basis for accurate goal-inference with partial and noisy observations of actual real-world demonstrations. We also show a grasp selection process, guided by the inferred human intention, to illustrate the use of the system for goal-directed grasp imitation.

Stochastic Optimization for Rigid Point Set Registration

In this paper we propose a new method for pairwise rigid point set registration. We pay special attention to noise robustness, outlier resistance and global optimal alignment. The problem of registering two point clouds in space is converted to a minimization of a nonlinear cost function. We propose a cost function that aims to reduce the impact of noise and outliers. Its definition is based on the input point sets and is directly related to the quality of a concrete rigid transform between them. In order to achieve a global optimal registration, without the need of a good initial alignment, we develop a new stochastic approach for global minimization. Tests on a variety of point sets show that the proposed registration algorithm performs very well on noisy, outlier corrupted and incomplete data.

A pilot study in vision-based augmented telemanipulation for remote assembly over high-latency networks

In this paper we present an approach to extending the capabilities of telemanipulation systems by intelligently augmenting a human operators motion commands based on quantitative three-dimensional scene perception at the remote telemanipulation site. This framework is the first prototype of the Augmented Shared-Control for Efficient, Natural Telemanipulation (ASCENT) System. ASCENT aims to enable new robotic applications in environments where task complexity precludes autonomous execution or where low-bandwidth and/or high-latency communication channels exist between the nearest human operator and the application site. These constraints can constrain the domain of telemanipulation to simple or static environments, reduce the effectiveness of telemanipulation, and even preclude remote intervention entirely. ASCENT is a semi-autonomous framework that increases the speed and accuracy of a human operators actions via seamless transitions between one-to-one teleoperation and autonomous interventions. We report the promising results of a pilot study validating ASCENT in a transatlantic telemanipulation experiment between The Johns Hopkins University in Baltimore, MD, USA and the German Aerospace Center (DLR) in Oberpfaffenhofen, Germany. In these experiments, we observed average telemetry delays of 200ms, and average video delays of 2s with peaks of up to 6s for all data. We also observed 75% frame loss for video streams due to bandwidth limits, giving 4fps video.

Blue-noise optimized point sets based on procrustes analysis

In this paper, we propose a new method for optimizing the blue noise characteristics of point sets. It is based on Procrustes analysis, a technique for adjusting shapes to each other by applying optimal elements of an appropriate transformation group. We adapt this technique to the problem at hand and introduce a very simple, efficient and provably convergent point set optimizer.

Local, Polynomial G 1 PN Quads

In this paper, we introduce the concept of local, polynomial G 1 PN quads. These are degree bi-5 polynomial surface patches in Bernstein-Bezier form. As the classic PN patch, ours interpolates the vertices of a quadrilateral control polygon and is orthogonal to a normal specified at each vertex. In contrast to the original concept, the proposed quad is orthogonal to four (continuous) normal fields — one defined at each boundary. Each of these normal fields and the corresponding patch boundary are uniquely determined by the data at two adjacent vertices of the control polygon. Thus, the patch construction is local in the sense that it is based solely on the information provided at the four control vertices. In this way, it is easy to stitch together multiple quads to construct a manifold G 1 continuous surface of arbitrary topological type. In contrast to other approaches, vertices at which 3 or more than 4 patches meet do not require special treatment.

Visualizing morphogenesis and growth by temporal interpolation of surface-based 3D atlases

Image-based 3D atlases have been proven to be very useful in biological and medical research. They serve as spatial reference systems that enable researchers to integrate experimental data in a spatially coherent way and thus to relate diverse data from different experiments. Typically such atlases consist of tissue-separating surfaces. The next step are 4D atlases that provide insight into temporal development and spatio- temporal relationships. Such atlases are based on time series of 3D images and related 3D models. We present work on temporal interpolation between such 3D atlases. Due to the morphogenesis of tissues during biological development, the topology of the non-manifold surfaces may vary between subsequent time steps. For animation therefore a smooth morphing between non-manifold surfaces with different topology is needed.