443Applications of Augmented Reality for the Automotive Industry
frame, and an arbitrary scale factor when a single camera is used. However, this last
drawback can be partially resolved by bootstrapping the 3D model of the scene with
the help of a model-based tracking approach. While the model-based solution is used
to estimate the pose of the rst keyframe, an initial reconstruction of the scene is
achieved by back-projecting the 2D image features onto the CAD model of the target
object (Bleser etal. 2006). If this initial reconstruction process provides a coordinate
frame and a scale factor to the VSLAM localization, their accuracy remains limited
since the reconstruction is achieved from a single point of view.
In the following, we wil