Affine Feature Matching via Stochastic Prediction

Tracking features from visual sensors for navigation purposes has emerged as a promising augmentation to convention sensors such as inertial measurement units (IMUs) and the Global Positioning System (GPS). In a tightly-coupled Extended Kalman Filter, errors may be reduced by approximately two orders of magnitude [11]. While the basic mathematics and algorithms are thoroughly documented, image-aided navigation is still in its early stages. This research improves image-aided navigation's feature tracking and landmark database by improving feature matching and landmark characterization across a wide range of viewpoints. In particular, current feature descriptors are typically based upon a Scale Invariant Feature Transform (SIFT) [5] and can only be matched reliably when viewed within +/- 30 degrees of the original viewpoint [1]. In this paper, it is shown experimentally that stochastic affine prediction expands the viewpoint validity to +/- 60 degrees. Furthermore, this description improves landmark databases by including a viewpoint dependency. Using real-world data, affine feature matching via stochastic prediction reduces navigation errors by 24% in position and 35% in attitude compared to the standard two-camera image-aided navigation setup.