Estimating motion with principal component regression strategies

TL;DR

This paper introduces two simple principal component regression methods for estimating optical flow in video sequences, offering robust, noise-agnostic alternatives that leverage local smoothness assumptions and simple image properties.

Contribution

The paper presents novel, easy-to-implement principal component regression strategies for optical flow estimation that do not require prior noise distribution knowledge.

Findings

Robust optical flow estimates achieved without noise distribution assumptions

Methods effectively handle mixtures of motion vectors

Preliminary experiments show promising results

Abstract

In this paper, two simple principal component regression methods for estimating the optical flow between frames of video sequences according to a pel-recursive manner are introduced. These are easy alternatives to dealing with mixtures of motion vectors in addition to the lack of prior information on spatial-temporal statistics (although they are supposed to be normal in a local sense). The 2D motion vector estimation approaches take into consideration simple image properties and are used to harmonize regularized least square estimates. Their main advantage is that no knowledge of the noise distribution is necessary, although there is an underlying assumption of localized smoothness. Preliminary experiments indicate that this approach provides robust estimates of the optical flow.