Employing Universal Voting Schemes for Improved Visual Place Recognition Performance

TL;DR

The paper addresses the problem of robust visual place recognition using ensemble VPR methods by systematically evaluating five universal voting schemes (Plurality, Condorcet, Broda Count, Contingent, IRV) across eight VPR techniques on six standard datasets. It introduces a framework that applies these voting rules to aggregate candidate matches, leveraging radar bounds, precision-recall curves, and a McNemar-like statistical test to assess significance. Results demonstrate strong dataset dependence, with Condorcet often delivering consistent improvements and IRV/Contingent voting excelling in specific contexts, challenging the assumption that a single voting rule is universally best. The findings provide practical guidance for selecting voting schemes in VPR ensembles and highlight the importance of statistical validation when comparing voting-based fusion approaches.

Abstract

Visual Place Recognition has been the subject of many endeavours utilizing different ensemble approaches to improve VPR performance. Ideas like multi-process fusion, Fly-Inspired Voting Units, SwitchHit or Switch-Fuse involve combining different VPR techniques together, utilizing different strategies. However, a major aspect often common to many of these strategies is voting. Voting is an extremely relevant topic to explore in terms of its application and significance for any ensemble VPR setup. This paper analyses several voting schemes to maximise the place detection accuracy of a VPR ensemble set up and determine the optimal voting schemes for selection. We take inspiration from a variety of voting schemes that are widely employed in fields such as politics and sociology and it is evident via empirical data that the selection of the voting method influences the results drastically. The paper tests a wide variety of voting schemes to present the improvement in the VPR results for several data sets. We aim to determine whether a single optimal voting scheme exists or, much like in other fields of research, the selection of a voting technique is relative to its application and environment. We propose a ranking of these different voting methods from best to worst which allows for better selection. While presenting our results in terms of voting method's performance bounds, in form of radar charts, PR curves to showcase the difference in performance and a comparison methodology using a McNemar test variant to determine the statistical significance of the differences. This test is performed to further confirm the reliability of outcomes and draw comparisons for better and informed selection a voting technique.

Figures (5)

• Figure 1: Sample output from the proposed experimental setup to evaluate the difference in performance bounds between each voting methodology. The red line in the radar chart is representative of the performance bounds (images correctly matched) and the axis represent the total number of query images in the data set. The various dimensions are useful to interpret the difference in performance of a voting scheme, in comparison to the others, such that the closer the red line is to boundary the better the performance of the voting scheme.
• Figure 2: A standard VPR ensemble set up employing several VPR methods simultaneously, that produces the top best matches by each method which are subjected to various voting schemes to observe difference in results.
• Figure 3: Difference in performance bounds of each voting methodology including Plurality, Condorcet, Contingent, Broda Count and Instant Run Off voting, in terms of query images correctly matched for different data sets : 17Places (top left), Livingroom (top right), Corridor (center left), CrossSeasons (center right), ESSEX3IN1 (bottom left) and GardensPoint (bottom right).
• Figure 4: Performance in terms of PR curves for each voting methodology including Plurality, Condorcet, Contingent Voting, Broda Count and Instant Run Off voting, in terms of query images correctly matched for different data sets : 17Places (top left), Livingroom (top right), Corridor (center left), CrossSeasons (center right), ESSEX3IN1 (bottom left) and GardensPoint (bottom right).
• Figure 5: Pairwise comparisons between the voting methods have been considered. A sign convention is used to present the results: a positive value of Z indicates that the first method of the pair outperforms the second one, whereas a negative Z score has the opposite meaning. Corresponding to the z-scores the legend represents the confidence intervals starting with the highest confidence intervals being green to lowest being red.