II. PROBLEM DESCRIPTION
The problem of Augmented Reality–enhanced structural
inspection path planning is, within the scope of this work,
defined as the challenge to fuse automatically generated optimized
inspection paths given a prior model of the inspection
structure S, with human head motion–based teleoperation
commands based on the AR–interface and the streamed live
feed and online 3D mapping data of the actual, real–life and
potentially different than S, structure S
′
. The 3D structure
to be inspected may be represented in a computationally
efficient way such as a triangular mesh or a octree–based
voxelgrid occupancy map (octomap) [21] and is embedded
into a bounded environment that may contain obstacle regions.
The problem setup is to be such that that for each
triangle in the mesh, there exists an admissible viewpoint
configuration – i.e. a viewpoint from which the triangle
is visible for a specific sensor model. Then, for the given
environment and with respect to the operational constraints,
a path connecting all viewpoints has to be found which
guarantees complete inspection of the 3D structure. Quality
measures for paths are situation specific, depending on the
system and mission objectives, e.g. shortest time or distance.
As a sample system we consider an aerial robot of the
Micro Aerial Vehicle (MAV) class equipped with a stereo
visual camera with a fixed orientation relative to the platform
body frame.