ArticlePDF Available

Informative Views and Sequential Recognition

Authors:
Article

Informative Views and Sequential Recognition

Abstract and Figures

In this paper we introduce a method for distinguishing between informative and uninformative viewpoints as they pertain to an active observer seeking to identify an object in a known environment. The method is based on a generalized inverse theory using a probabilistic framework where assertions are represented by conditional probability density functions. Consequently, the method also permits the assessment of the beliefs associated with a set of assertions based on data acquired from a particular viewpoint. The importance of this result is that it provides a basis by which an external agent can assess the quality of the information from a particular viewpoint, and make informed decisions as to what action to take using the data at hand. To illustrate the theory we show how the characteristics of belief distributions can be exploited in a model-based recognition problem, where the task is to identify an unknown model from a database of known objects on the basis of parameter estimates. This leads to a sequential recognition strategy in which evidence is accumulated over successive viewpoints (at the level of the belief distribution) until a definitive assertion can be made. Experimental results are presented showing how the resulting algorithms can be used to distinguish between informative and uninformative viewpoints, rank a sequence of images on the basis of their information (e.g. to generate a set of characteristic views), and sequentially identify an unknown object. 1 R'esum'e Dans cet article, nous pr'esentons une m'ethode qui permet de distinguer les points de vue informatifs et non-informatifs d'un objet tels que percus par un observateur actif qui cherche `a identifier un objet dans un environnement connu. La m'ethode repose sur une g'en'eralisation de la...
Content may be subject to copyright.
(a) (b) (c)
Big Sphere
Block
Cylinder
Lemon
Small Sphere
Round Block
-40
-30
-20
-10
0
U
Log Belief
Round Block
Small Sphere
Lemon
Cylinder
Block
Big Sphere
True Parts
B
Bs
C
L
EarL EarR
EyeL EyeR
Head Nose Rb
Ss
-40
-30
-20
-10
U
Log Belief
Nose
Head
EyeR
EyeL
EarR
EarL
Cap
True Parts
H N ERL ERR EYL EYR
2.5
5
7.5
10
12.5
15
Undetermined
Incorrect
Correct
Number of Trials
H N ERL ERR EYL EYR
5
10
15
20
25
30
Undetermined
Incorrect
Correct
Number of Trials
... In order to perform efficient reconstruction that maximises quality and coverage using a minimum amount of data (such as in [1] and [24]), it is necessary to actively select where the NBV is. NBV estimation can be divided into two main categories: exploration and refinement. ...
... This novel formulation allows us to estimate paths with Algorithm 1 RRT version of Scenic Pathplanner. 1: function BUILDSCENICRRT(x init ,x goal ) 2: G.ADD VERTEX(x init ) 3: ...
... Whaite and Ferrie [12] presented an autonomous exploration theory based on minimization of model uncertainty. Callari and Ferrie [13] and Arbel and Ferrie [14,15] propose a method for accumulating evidence in the form of probability distributions and entropy measures in the framework of active recognition. And more recently, Zhou and Leung [16] have presented the minimun entropy fusion approach for multisource data fusion in non-Gaussian processes. ...
... The idea of aspect graph (conceptually very similar to the viewing representation) was proposed by Koenderink and van Doorn [2]. Arbel , and Ferrie [1] introduce a method for distinguishing between informative and uninformative viewpoints from views are obtained. In this paper we deal with the viewer-centered category. ...
Article
Full-text available
This article provides relationship between scanning resolution of acquisition device and flare of angle of viewing cone. Knowing the technical parameters of equipment, a viewing representation adapted to this device can be generated. The representation of polyhedral object is obtained from viewing sphere with perspective model.
... 2 1 2 D view also contains volumetric information about the visible part of the general model. Application of 2 1 2 D views instead of images (2D views) eliminates the need of building a views neighbourhood structure like "aspect graphs" ([2, 6, 16]), or performing a "next view planning" procedure ( [8,10,14]). In the paper [11], a method of feature dependent viewpoints location on the view sphere using the perspective projection is developed and suitable procedure of views generation is described. ...
Article
Full-text available
The paper concerns the construction of a complete view representation of 3D exact models of objects for a model-based visual identification system. The system uses 2 1 2 D views and depth map of the scene. In particular, a new method of generating convex polyhedra views is described. The views are of the 2 1 2 D type, exact, and are made in a perspective projection. The method is based on a new concept of the view sphere with perspective. In particular, a concept of a completeness of a set of views in a view model is presented. The model is generated in a loop: generating the (distinct) feature dependent views → determining the single-view areas → finding the gaps (holes) in the view sphere covering (by the single-view areas) → generating the extra (gaps) views for filling the gaps → determining single-view areas for the views → returning to the gaps finding. The loop ends when the whole view sphere is covered by the single-view areas. A method of determining single-view areas for convex polyhedra is also presented. Keywords: visual object identification, imaging geometry, view sphere with perspective, geometric models, standarized views, 2 1 2 D complete view model, distinct features.
... That why overall computational complexity of this algorithm should by considered as closest to n 2 than to n 4 . 6 ...
Article
Full-text available
This article concerns generating of 3D multiview model of non convex polyhedron, according to viewing sphere with perspective projection. Those models are going to be used for visual identification based on them and a scene depth map. We give a new concept and an algorithm for face-depended generation of multiview model. It does not require any preprocessing nor auxiliary mechanisms or complex calculations connected with them.
Article
Full-text available
This paper is the second part of our study describing methods for obtaining a 3D multiview exact and complete model of convex polyhedra used for visual identification. Non-iterative methods, like the iterative ones [21], explore the concept of the view sphere with perspective projection and the view sphere covering as a mechanism for representation completeness.The non-iterative methods ([19]) consist in calculating a view, determining the corresponding single-view area (the so-called seedling single-view area) and then searching for the neighbouring single-view areas (generating the views at the same time) in a spiral way until the whole view sphere is covered by the latter. Having a complete set of the single-view areas (complete view sphere covering), we get a complete set of views as well.Test results and computational complexity estimation are also included.
Conference Paper
Full-text available
This article concerns generating of 3D multiview model of convex polyhedron that are a complete representation of this polyhedron, according to viewing sphere with perspective projection. Those models are going to be used for visual identification based on them and a scene depth map. We give a new concept and an algorithm for face-depended generation of multi-face views. It does not require any preprocessing nor auxiliary mechanisms or complex calculations connected with them.
Chapter
Full-text available
This article concerns generating 3 D multiview models of convex polyhedral that are a complete representation of this polyhedral, according to viewing sphere with perspective concept. Those models are going to be used for visual identification based on them and a scene depth map. We give a new conception and an algorithm for face-depended generation of multi-face views. It does not require any preprocessing nor auxiliary mechanisms or complex calculations connected with them.
Article
This article deals with the construction of a complete 2(1/2)D exact view models of polyhedral objects for visual identification systems. In particular, a new method and an algorithm of views generation using the {\bf view sphere with perspective} concept is described. A set of views generated by this method forms a complete view representation of the object. The method of ensuring completeness of the view representation by controlling covering of a view space (by the single-view areas) is used in presented algorithm. The perspective projection used for calculating the views, the total, tightly covering of the view sphere by the single-view areas and 2(1/2)-dimensionality of the views secure, in our opinion, unambiguous and proper identification of polyhedral objects. The method consists in calculating one (any) view, determining corresponding single-view area (so called seedling single-view area) and then spiral multiplying neighbouring single-view areas untill the whole view sphere is covered by them (i.e. untill the border register containing the border between covered and uncovered parts of the view sphere occures empty). Having a complete set of single-view areas we are getting a complete set of views as well. A method of determining single-view areas for convex polyhedra is also presented.
Article
Full-text available
This paper is the second part of our study describing methods for obtaining a 3D multiview exact and complete model of convex polyhedra used for visual identification. Non-iterative methods, like the iterative ones [21], explore the concept of the view sphere with perspective projection and the view sphere covering as a mechanism for representation completeness. The non-iterative methods ([19]) consist in calculating a view, determining the corresponding single-view area (the so-called seedling single-view area) and then searching for the neighbouring single-view areas (generating the views at the same time) in a spiral way until the whole view sphere is covered by the latter. Having a complete set of the single-view areas (complete view sphere covering), we get a complete set of views as well. Test results and computational complexity estimation are also included.
Article
Full-text available
This paper presents a comparative study and survey of model-based object-recognition algorithms for robot vision. The goal of these algorithms is to recognize the identity, position, and orientation of randomly oriented industrial parts. In one form this is commonly referred to as the "bin-picking" problem, in which the parts to be recognized are presented in a jumbled bin. The paper is organized according to 2-D, 2½-D, and 3-D object representations, which are used as the basis for the recognition algorithms. Three central issues common to each category, namely, feature extraction, modeling, and matching, are examined in detail. An evaluation and comparison of existing industrial part-recognition systems and algorithms is given, providing insights for progress toward future robot vision systems.
Article
We define autonomous exploration as a process by which an active observer can interact with its surroundings, e.g., by intentionally moving around and collecting information, in order to learn about its environment. Such an ability is essential for autonomous systems that must operate in unstructured environments, i.e., where it is difficult (if not impossible) to characterize the environment beforehand. This paper describes a working system that implements an autonomous explorer whose function is to describe the environment in terms of articulated volumetric models. A novel feature of the system is that it uses feedback to reduce reliance on a priori knowledge.
Article
The book provides an up-to-date description of the methods used for fitting experimental data, or to estimate model parameters, and to unify these methods into the Inverse Problem Theory. The first part of the book deals with problems and describes Maximum likelihood, Monte Carlo, Least squares, and Least absolute values methods. The second part deals with inverse problems involving functions. Theoretical concepts are emphasized, and the author has all the useful formulas listed, with many special cases included. The book serves as a reference manual.
A method is presented for partitioning a set of surface estimates obtained with a laser range finding system into subsets corresponding to parts of an object. The authors' strategy uses two complementary representations for surfaces, i.e., one which describes local structures in terms of differential properties (e.g., edges, lines, contours) and another which represents the surface as a collection of smooth patches at different scales. By enforcing a consistent interpretation between these two representations, it is possible to derive a partitioning algorithm that is both efficient and robust
Article
Several researchers have recently addressed the problem of computing the aspect graph of an object by deriving the exact partition of viewpoint space from the object's geometry. Algorithms have been developed to handle general polyhedral objects, with viewpoint space modeled as either the Gaussian sphere or as 3-D space. This paper presents the first algorithm to handle a defined class of curved-surface objects. The algorithm automatically computes the partition of the Guassian sphere, and thereby the aspect graph, for solids of revolution defined as Right, Circular, Straight, Homogeneous Generalized Cylinders.
Conference Paper
The authors formulate the segmentation task as a search for a set of descriptions which minimally encodes a scene. A novel framework for cooperative robust estimation is used to estimate descriptions that locally provide the most savings in encoding an image. A modified Hopfield-Tank networks finds the subset of these descriptions which best describes an entire scene, accounting for occlusion and transparent overlap among individual descriptions. Using a part-based 3-D shape model the authors have implemented a system that is able to successfully segment images into their constituent structure
Article
This paper introduces a new approach to computing the exact orthographic aspect graph of curved objects. Curves corresponding to various visual events partition the view sphere into regions where the image structure is stable. A catalog of these events for piecewise-smooth objects is available from singularity theory. For a solid of revolution whose generator is an algebraic curve, each visual event is characterized by a system of polynomial equations whose roots can be computed by continuation methods. Within each region, the stable image structure is characterized by a variation of cylindrical algebraic decomposition and ray tracing. This approach has been implemented and several examples are presented.
Article
This paper attempts to relate two influential families of 3D shape primitives, superquadrics and geons. Superquadric surfaces have attracted considerable interest recently as part models for representation of 3D objects. Here, we explore their utility for a different purpose, i.e. for shape discrimination. We investigate how well the estimated superquadric parameters can reflect certain intuitive geometric attributes of elongated objects, such as axis shape (straight or bent), type of cross-sectional edges (straight or curved), and variation of cross-section size along the axis (constant, tapered, or increasing-and-decreasing). Parameters of the best-fitting superquadric are estimated for real as well as synthetic range images, obtained from a large number of viewpoints, of models belonging to 12 shape classes based on the above geometric attributes. These shape classes correspond to a ‘collapsed’ set of 36 different geons. Five features derived from the estimated superquadric parameters are used to distinguish between these 12 shape classes. Classification error rates are estimated for k-nearest-neighbour and binary tree classifiers. The effects of varying the range image resolution, noise level and the model elongation are also investigated. The results indicate that existing methods of superquadric parameter estimation are quite sensitive to noise in range values and also do not work well for ‘rough’, coarse-surfaced objects. However, for low-noise real range images of objects with ‘smooth’ surfaces, numerical features derived from estimated superquadric parameters can be used with a binary tree classifier to infer qualitative shape properties with about 80% reliability.
Article
The goal in computer vision systems is to analyze data collected from the environment and derive an interpretation to complete a specified task. Vision system tasks may be divided into data acquisition, low-level processing, representation, model construction, and matching subtasks. This paper presents a comprehensive survey of model-based vision systems using dense-range images. A comprehensive survey of the recent publications in each subtask pertaining to dense-range image object recognition is presented.