Sequence-Based Filtering for Visual Route-Based Navigation: Analyzing the Benefits, Trade-Offs and Design Choices

Visual Place Recognition (VPR) is the ability to correctly recall a previously visited place using visual information under environmental, viewpoint and appearance changes. An emerging trend in VPR use of sequence-based filtering methods on top single-frame-based matching techniques for route-based navigation. The combination leads varying levels potential performance boosts at increased computational costs. This raises number interesting research questions: How does boost (due sequential filtering) vary along entire spectrum methods? sequence length affect curve? Which specific combinations provide good trade-off between computation? However, there lack previous work looking these important questions most date has been used without systematic approach. To bridge this gap, paper conducts an in-depth investigation relationship those methods. It analyzes individual trade-offs, properties limitations different techniques. experiments conducted study demonstrate benefits over approach various We found that applying lightweight descriptor can enable higher accuracy than state-of-the-art such as NetVLAD, while running shorter time. For example, 16 images, CALC outperforms NetVLAD Campus Loop dataset taking about 22% less time perform VPR.