ConvSequential-SLAM: A Sequence-based, Training-less Visual Place Recognition Technique for Changing Environments

Visual Place Recognition (VPR) is the ability to correctly recall a\npreviously visited place under changing viewpoints and appearances. A large\nnumber of handcrafted and deep-learning-based VPR techniques exist, where the\nformer suffer from appearance changes and the latter have significant\ncomputational needs. In this paper, we present a new handcrafted VPR technique\nthat achieves state-of-the-art place matching performance under challenging\nconditions. Our technique combines the best of 2 existing trainingless VPR\ntechniques, SeqSLAM and CoHOG, which are each robust to conditional and\nviewpoint changes, respectively. This blend, namely ConvSequential-SLAM,\nutilises sequential information and block-normalisation to handle appearance\nchanges, while using regional-convolutional matching to achieve\nviewpoint-invariance. We analyse content-overlap in-between query frames to\nfind a minimum sequence length, while also re-using the image entropy\ninformation for environment-based sequence length tuning. State-of-the-art\nperformance is reported in contrast to 8 contemporary VPR techniques on 4\npublic datasets. Qualitative insights and an ablation study on sequence length\nare also provided.\n