InInVi originates from ERC Project VarCity. VarCity aims at generating 3D city models that include the city
buzz, like traffic. Another such dynamic aspect in VarCity is the enhancement with interesting videos, AR
(Augmented Reality) style: this “AR Nouveau” functionality will show automatically generated compilations
of user videos overlaid on the 3D city models. Such overlay requires the estimated camera motion to be very precise.
Thus, current AR applications make use of dedicated acceleration sensors and electronic gyroscopes
built in in nowadays mobile phones or head up displays. In the VarCity-Project, we reached a breakthrough in the analysis of camera motion in user-
provided videos where these sensors are not available. This enables highly realistic AR applications in fields
where the aforementioned additional sensors are not available, e.g. AR in live television.
Based on these findings, in this proof of concept (PoC) project, we analysed the market for sports TV
advertisements in different sports and how to best adapt VarCity’s camera motion extraction to access it with
maximal impact. We developed a PoC system which automatically analyses camera motions of one or multiple
cameras in live and post-produced television signals with very high accuracy and robustness.
This process is a key step towards virtual advertising, where the
underlying idea is to replace the classical world feed distribution by individualised unique
television feeds or streams to allow for new applications and advertisement concepts.
