<title>Feature space trajectory representation for active vision</title>

Abstract

A new feature space trajectory (FST) description of 3-D distorted views of an object is advanced for active vision applications. In an FST, different distorted object views are vertices in feature space. A new eigen-feature space and Fourier transform features are used. Vertices for different adjacent distorted views are connected by straight lines so that an FST is created as the viewpoint changes. Each different object is represented by a distinct FST. An object to be recognized is represented as a point in feature space; the closest FST denotes the class of the object, and the closest line segment on the FST indicates its pose. A new neural network is used to efficiently calculate distances. We discuss its uses in active vision. Apart from an initial estimate of object class and pose, the FST processor can specify where to move the sensor to: confirm class and pose, to grasp the object, or to focus on a specific object part for assembly or inspection. We advance initial remarks on the number of aspect views needed and which aspect views are needed to represent an object.

Global feature space neural network for active computer vision

“…where the PDF for x conditioned on class is obtained by multiplying the PDF for x conditioned on class and pose by the PDF for the pose parameter and integrating over t9 19) po (19)dO (7) The unconditional PDF for x, p,, (x) in Eq. (6), is obtained using the total probability theorem [46] as…”

“…We consider the FST representation of 3D distorted views of an object as a component of active vision system [7] that uses data from simple and inexpensive sensors (2D CCD cameras vs. 3D range sensors). In this paper, we describe new algorithms using the FST representation and neural network processor, that determine the best object viewpoints to use to improve the class decision or pose estimates.…”

Untitled

“…Optic flow has been applied by Shirai et al [31], Allen et al [1] and Papanikolopoulos [27]. Sipe et al [32], Nayar et al [24] and Deguichi [6] used eigenfeatures for 3-dof positioning and tracking.…”

“…In the field of computer vision, Principal Component Analysis (PCA) is best known as a method for image compression, feature detection, and pattern recognition. Recently, however, several authors have demonstrated its usefulness for pose estimation [24,32,6]. Given a set of multidimensional data samples, the aim of PCA is to determine a reduced orthonormal basis whose axes are oriented in the directions of maximum variance of the data in each dimension, and then project the data onto this new basis.…”

Global feature space neural network for active object recognition

“…2). The ability of the FST to utilize a reduced number of object views gives it a major advantage over other distorted object representations: whether the FST is constructed from a prototype object or a computer aided design (CAD) model, a continuum of training views is possible and some means of selecting views is necessary [5,6].…”