Unsupervised Place Recognition with Deep Embedding Learning over Radar Videos

TL;DR

This paper introduces an unsupervised deep learning method to generate embeddings from radar videos for place recognition, outperforming supervised methods on large-scale data with high accuracy.

Contribution

It presents a novel unsupervised learning approach for radar-based place recognition that achieves state-of-the-art performance without labeled data.

Findings

Achieved 98.38% localization accuracy on 280 km of radar data.

Outperformed existing supervised place recognition methods.

Demonstrated effectiveness of unsupervised deep embeddings for complex radar data.

Abstract

We learn, in an unsupervised way, an embedding from sequences of radar images that is suitable for solving place recognition problem using complex radar data. We experiment on 280 km of data and show performance exceeding state-of-the-art supervised approaches, localising correctly 98.38% of the time when using just the nearest database candidate.