Bayesian Weighted Summation of the Absolute Difference and Its Implications for Robust Visual Navigation

Abstract

Visual navigation has a long history of taking inspiration from biology, owing to the wide range of applications facilitated by biological vision, such as feature detection and navigation. Of capital interest is the ability of small-brained organisms, such as bees and sub-Saharan ants, to navigate using dynamically changing landmarks and scenery. Biological research literature suggests that bees and ants can achieve this impressive feat by navigating toward scenes that are familiar. By contrast, the capability of computationally simple vision-based robotic methods, such as sum of the absolute difference (SAD), exhibit significantly degraded performance in changing environments. This disparity is addressed in this work, through a novel approach which determines regions within an image that are robust to environmental change by using the weighted sum of the absolute difference (BWSAD). By utilizing these regions, BWSAD successfully navigates the paths in changing environments at a higher rate than the SAD baseline.