Recognition and pose estimation of unoccluded three-dimensional objects from a two-dimensional perspective view by banks of neural networks

Abstract

This paper describes a neural network (NN) based system for recognition and pose estimation of an unoccluded three-dimensional (3-D) object from any single two-dimensional (2-D) perspective view. The approach is invariant to translation, orientation, and scale. First, the binary silhouette of the object is obtained and normalized for translation and scale. Then, the object is represented by a set of rotation invariant features derived from the complex orthogonal pseudo-Zernike moments of the image. The recognition scheme combines the decisions of a bank of multilayer perceptron NN classifiers operating in parallel on the same data. These classifiers have different topologies and internal parameters, but are trained on the same set of exemplar perspective views of the objects. Next, two pose parameters, elevation and aspect angles, are obtained by a novel two-stage NN system consisting of a quadrant classifier followed by NN angle estimators. Performance is tested on clean and noisy data bases of military ground vehicles. Comparative studies with three other classifiers (a single NN, the weighted nearest-neighbor classifier, and a binary decision tree) are carried out.

Application of multi-layer perceptron neural networks to vision problems

“…A multi-MLP classification system is employed [6]. This system includes several different MLP classifiers (three in this study) that operate in parallel on the same set of invariant PZM features.…”

“…The first problem is 3D target recognition and pose estimation from a single 2D view (summarised from Khotanzad and Liou [6]); and the second is a new approach to recognition of handwritten digits.…”

Learning multiview face subspaces and facial pose estimation using independent component analysis

“…Khotanzad and Liou [26] represent three-dimensional objects by a set of rotation invariant features derived from the complex orthogonal pseudoZernike moments of their two-dimensional (2-D) perspective images, and then obtain the pose parameters, i.e., aspect and elevation angles of the objects, by a two-stage neural network system.…”

Colour-based rigid body tracking using three-dimensional graphics models

“…Model-based rigid body tracking methods may use edge features (Li & Wang 1999), point features or region features (Khotanzad & Liou 1996). Hel-Or & Werman (1995) used a feature point model-based approach to locate a planar object in four different poses to within 0.009 m and 1.1° of translation and rotation respectively, with ranges of 0.49 m and 46.0° respectively.…”