Histogram of Oriented Cameras - A New Descriptor for Visual SLAM in Dynamic Environments

Abstract

Simultaneous localization and mapping (SLAM) is a basic prerequisite in autonomous mobile robotics. Most existing visual SLAM approaches either assume a static environment, or simply ’forget’ old parts of the map to cope with map size constraints and scene dynamics. We present a novel map representation for sparse visual features. A new 3D point descriptor called Histogram of Oriented Cameras (HOC) encodes anisotropic spatial visibility information and the importance of each three-dimensional landmark. Each feature holds and updates a histogram of the poses of observing cameras. It is hereby able to estimate its probability of occlusion and importance for localization from a given viewpoint. In a series of simulated and real-world experiments we prove that the proposed descriptor allows to cope with dynamic changes in the map, improves localization accuracy and enables reasonable control of the map size.