Static Parameter Estimation on SO(3) using Stochastic Gradient Descent for Visual Tracking

This paper presents a method for parameter estimation using a particle filter, where the parameters are elements on the manifold SO(3). The presented method is based on Stochastic Gradient Descent, which makes it possible to perform the static parameter estimation on-line. The presented method is coordinate-free and is therefore not prone to singularities. The proposed method is demonstrated for estimating the motion of a crane load, where markers on the crane load facilitate visual tracking. The pose of each marker relative to the crane load is initially unknown, but are found within four seconds using the proposed method.