Probabilistic Visual Place Recognition for Hierarchical Localization

Abstract

Visual localization techniques often comprise a hierarchical localization pipeline, with a visual place recognition module used as a coarse localizer to initialize a pose refinement stage. While improving the pose refinement step has been the focus of much recent research, most work on the coarse localization stage has focused on improvements like increased invariance to appearance change, without improving what can be loose error tolerances. In this letter, we propose two methods which adapt image retrieval techniques used for visual place recognition to the Bayesian state estimation formulation for localization. We demonstrate significant improvements to the localization accuracy of the coarse localization stage using our methods, whilst retaining state-of-the-art performance under severe appearance change. Using extensive experimentation on the Oxford RobotCar dataset, results show that our approach outperforms comparable state-of-the-art methods in terms of precision-recall performance for localizing image sequences. In addition, our proposed methods provides the flexibility to contextually scale localization latency in order to achieve these improvements. The improved initial localization estimate opens up the possibility of both improved overall localization performance and modified pose refinement techniques that leverage this improved spatial prior.

Probabilistic Appearance-Invariant Topometric Localization With New Place Awareness

“…Xu et al [10] adapt appearance-invariant image descriptors like NetVLAD [4], DenseVLAD [1] and APGeM [5] to the Bayesian state-estimation framework, demonstrating state-ofthe-art performance for global localization with substantial appearance change between query and reference images. The authors provide both a discrete topological filter that does not use odometry information, and a particle filter that utilizes odometry.…”

“…The measurement model scores the likelihood of a query image given a particular robot state p(z q t |x t ). For within-map states, we follow the method proposed in [10], where the likelihood for state x t is given by an exponential kernel over the appearance signatures…”

“…where λ is calibrated using z q 0 as in [10]. The off-map likelihood is set as the likelihood of the k th best match for k >> 1.…”

“…Global localization involves localizing a robot to a map assuming no prior information is available around its location. Xu et al [10] proposed a method for addressing the wakeup task using probabilistic filtering techniques, where the robot is first initialized with a uniform prior belief. Forward filtering steps are then applied given a query sequence until the belief has sufficiently converged around a specific place within the map; see [10] for detailed methodology.…”

“…Key to this system is a novel, principled technique for conditioning transition probabilities between discrete states with continuous 3-dof odometry. This extends existing discrete filters which use none [10], [17] to limited [9] odometry and provides an alternative to continuous state systems utilizing particle filters [10], [18].…”