PoleStack: Robust Pole Estimation of Irregular Objects from Silhouette Stacking

2025-02-05

AI Summary

This work addresses the challenging problem of estimating the principal rotation axis from multi-view silhouette images of irregular objects—particularly under conditions of unknown centroid image positions, severe surface shadowing, and centroid registration errors. We propose a robust method for principal axis direction estimation. Its core innovation lies in the first integration of silhouette stacking with Fourier magnitude spectrum analysis to construct a stacked image exhibiting reflection symmetry, enabling translation-invariant symmetry axis detection. Subsequently, multi-view geometric fusion is employed to recover the 3D orientation of the principal axis. Crucially, the method does not require precise centroid registration, significantly enhancing robustness against surface shadows, image noise, and low-resolution inputs. Experimental results demonstrate degree-level accuracy even under low-resolution conditions, validating its suitability for on-orbit pose estimation of non-cooperative targets during spacecraft proximity operations and hovering phases.

Technology Category

Intelligent Robots: State EstimationComputer Vision: Scene Analysis & UnderstandingMachine Learning: Multi-instance/Multi-view Learning

Application Category

Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphsSecurity and Privacy: Large-scale security measurementsWeb Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

We present an algorithm to estimate the rotation pole of a principal-axis rotator using silhouette images collected from multiple camera poses. First, a set of images is stacked to form a single silhouette-stack image, where the object's rotation introduces reflective symmetry about the imaged pole direction. We estimate this projected-pole direction by identifying maximum symmetry in the silhouette stack. To handle unknown center-of-mass image location, we apply the Discrete Fourier Transform to produce the silhouette-stack amplitude spectrum, achieving translation invariance and increased robustness to noise. Second, the 3D pole orientation is estimated by combining two or more projected-pole measurements collected from different camera orientations. We demonstrate degree-level pole estimation accuracy using low-resolution imagery, showing robustness to severe surface shadowing and centroid-based image-registration errors. The proposed approach could be suitable for pole estimation during both the approach phase toward a target object and while hovering.

Problem

• Estimate rotation pole from silhouette images
• Handle unknown center-of-mass image location
• Achieve robust 3D pole orientation estimation

Innovation

• Silhouette stacking for symmetry detection
• Discrete Fourier Transform for noise robustness
• Multi-camera orientation for 3D pole estimation