HM4: Hidden Markov Model with Memory Management for Visual Place Recognition

Visual place recognition needs to be robust against appearance variability\ndue to natural and man-made causes. Training data collection should thus be an\nongoing process to allow continuous appearance changes to be recorded. However,\nthis creates an unboundedly-growing database that poses time and memory\nscalability challenges for place recognition methods. To tackle the scalability\nissue for visual place recognition in autonomous driving, we develop a Hidden\nMarkov Model approach with a two-tiered memory management. Our algorithm,\ndubbed HM$^4$, exploits temporal look-ahead to transfer promising candidate\nimages between passive storage and active memory when needed. The inference\nprocess takes into account both promising images and a coarse representations\nof the full database. We show that this allows constant time and space\ninference for a fixed coverage area. The coarse representations can also be\nupdated incrementally to absorb new data. To further reduce the memory\nrequirements, we derive a compact image representation inspired by Locality\nSensitive Hashing (LSH). Through experiments on real world data, we demonstrate\nthe excellent scalability and accuracy of the approach under appearance changes\nand provide comparisons against state-of-the-art techniques.\n