Improving Visual Place Recognition Performance by Maximising Complementarity

Abstract

Visual place recognition (VPR) is the problem of recognising a previously visited location using visual information. Many attempts to improve the performance of VPR methods have been made in the literature. One approach that has received attention recently is the multi-process fusion where different VPR methods run in parallel and their outputs are combined in an effort to achieve better performance. The multi-process fusion, however, does not have a well-defined criterion for selecting and combining different VPR methods from a wide range of available options. To the best of our knowledge, this paper investigates the complementarity of state-of-the-art VPR methods systematically for the first time and identifies those combinations which can result in better performance. The letter presents a well-defined framework which acts as a sanity check to find the complementarity between two techniques by utilising a McNemar's test-like approach. The framework allows estimation of upper and lower complementarity bounds for the VPR techniques to be combined, along with an estimate of maximum VPR performance that may be achieved. Based on this framework, results are presented for eight state-of-the-art VPR methods on ten widely-used VPR datasets showing the potential of different combinations of techniques for achieving better performance.

SwitchHit: A Probabilistic, Complementarity-Based Switching System for Improved Visual Place Recognition in Changing Environments

“…[11] presents a three-tier hierarchical multi-process fusion system which is customizable and may be extended to any arbitrary number of tiers. Another interesting idea that explores the notion of complementarity between multiple VPR techniques is introduced by [12]. A McNemar's test like approach is used to test out the level of complementarity between different VPR techniques.…”

“…Rather than another attempt to develop a new VPR technique from scratch, a well received and an intuitive solution has been put forward in [10], [11] that introduces the concept of multi-process fusion between different VPR techniques. Furthermore, to refine the idea of multi-process fusion, [12] proposes a framework evaluating the complementarity of different VPR techniques with each other. This is based on a functional piece of knowledge collected through empirical data that different techniques have the potential of complementing each other's weaknesses.…”

“…This informed, instead of a random, selection of VPR techniques can said to be a key factor when developing a multi-process fusion system. Although [12] puts forward a framework and detailed analysis of the complementarity many state-of-the-art VPR techniques share on different data sets, it passively evaluates these different techniques without providing any real time evidence or results. In this paper we are introducing a system that attempts to tackle these shortcomings using the Bayes inference, to predict the probability a primary VPR technique, in a combined system, has for correctly matching the query image.…”

Hierarchical Multi-Process Fusion for Visual Place Recognition

“…More recently, evidence has been accumulating that certain combinations of modalities are more effective than others on specific datasets [6], [12]. However, both these works rely on either a training set, or access to the full ground-truth information for a given dataset.…”

“…By using multiple VPR techniques simultaneously, limitations with one technique (in a specific environment) can be offset with the inclusion of additional techniques that may not have such limitations. More recent work has shown that certain combinations of techniques are more effective than others on specific datasetsa phenomenon that has been termed the complementarity of different combinations of techniques [6], [12].…”