Optimizing Through Learned Errors for Accurate Sports Field Registration

TL;DR

This paper introduces a novel optimization framework that trains a deep network to regress registration errors, enabling more accurate sports field registration in broadcast videos by minimizing predicted errors.

Contribution

It presents a new method that combines deep learning with optimization for improved registration accuracy, surpassing existing state-of-the-art techniques.

Findings

Outperforms current state-of-the-art in real-world sports videos

Effective on synthetic toy examples with unlimited data

Significant gains in simplified registration scenarios

Abstract

We propose an optimization-based framework to register sports field templates onto broadcast videos. For accurate registration we go beyond the prevalent feed-forward paradigm. Instead, we propose to train a deep network that regresses the registration error, and then register images by finding the registration parameters that minimize the regressed error. We demonstrate the effectiveness of our method by applying it to real-world sports broadcast videos, outperforming the state of the art. We further apply our method on a synthetic toy example and demonstrate that our method brings significant gains even when the problem is simplified and unlimited training data is available.