Towards BAFOS: Bundle Adjustment with Feature Orientation and Scale

摘要

This paper proposes a new approach for enhanced-accuracy visionbasedpose estimation in GPS-denied unknown environments. The problemis typically addressed using bundle adjustment (BA) approaches,that minimize the difference between the projected and measured imagefeatures (re-projection error). We propose to incorporate within bundleadjustment new type of constraints that relate between image featurescale and orientation between matched views. While feature scale andorientation play an important role in image matching, they have notbeen utilized thus far for estimation purposes in BA framework. Our approachexploits feature scale and orientation information, that is alreadyavailable from the image matching process, and uses it to enhance theaccuracy of bundle adjustment. In this paper we make the first stepsin this direction and develop the corresponding feature scale constraint,which relates between image measured scale and the actual scale of theobserved landmark. We demonstrate the proposed concept in a statisticalsimulation.