B1319
Title: Some recent advances on regression models in shape data
Authors: Alfred Kume - University of Kent (United Kingdom) [presenting]
Abstract: Shape data are naturally represented as points in non-Euclidean spaces due to the natural process of their generation and the inherited invariances that need to be applied during the inferential process. For example, rotation invariance imposes a nonlinear constraint on the shape observations, which can typically be a collection of some landmark coordinates in 2 or 3-dimensional spaces. The general approaches with regression models applied to such data are listed. Among all the approaches developed so far, there has not been much focus on the generality of landmark correlations. This is due to the fact that a general covariance structure leads to identifiability issues with the respective parameters. We aim to tackle this problem by implementing a polynomial regression for modelling the data generated from mediapipe which provides automatic landmark extraction of the hand.
