Observability/Identifiability of Rigid Motion under Perspective Projection

Abstract

The "visual motion" problem consists of estimating the motion of an object viewed under projection. In this paper we address the feasibility of such a problem. We will show that the model which defines the visual motion problem for feature points in the euclidean 3D space lacks of both linear and local (weak) observability. The locally observable manifold is covered with three levels of lie differentiations. Indeed, by imposing metric constraints on the state-space, it is possible to reduce the set of indistinguishable states. We will then analyze a model for visual motion estimation in terms of identification of an Exterior Differential System, with the parameters living on a topological manifold, called the "essential manifold", which includes explicitly in its definition the forementioned metric constraints. We will show that rigid motion is globally observable/identifiable under perspective projection with zero level of lie differentiation under some general position conditions. Such conditions hold when the viewer does not move on a quadric surface containing all the visible points.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Mar 08, 1994
Accession Number
ADA464761

Entities

People

  • Pietro Perona
  • Stefano Soatto

Organizations

  • California Institute of Technology

Tags

Communities of Interest

  • Air Platforms
  • Sensors

DTIC Thesaurus Topics

  • California
  • Dynamics
  • Equations
  • Estimators
  • Filters
  • Identification
  • Kalman Filters
  • Noise
  • Nonlinear Dynamics
  • Nonlinear Systems
  • Notation
  • Observers
  • Relative Motion
  • Rotation
  • Statistical Analysis
  • Statistical Estimation
  • Translations

Fields of Study

  • Computer science

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Computer Vision.
  • Graph Algorithms and Convex Optimization.

Technology Areas

  • Space
  • Space - Space Objects