Improving Visual Place Recognition Performance by Maximising Complementarity

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