Enhanced Stereo Vision Using Free-Form Surface Mirrors

Abstract

In the domain of stereo vision, the presence of repetitive patterns results in multiple matching hypotheses. The choice of the wrong hypothesis leads to an incorrect distance measurement. In applications such as automotive vision-based navigation a high precision in matching and distance calculation is vital. A common approach is the use of multiple cameras. Unfortunately, in vehicle applications this is often not feasible. However, the shiny varnished body parts of the car supply a free-form surface mirror. In combination with a camera system they form virtual cameras with a different viewing direction. This additional information can be used to select the correct matching hypothesis and to increase the depth measurement accuracy. The free-form surface mirrors yield distorted pictures without a single viewpoint which prevents a purely perspective reconstruction. We will discuss the problems arising from the use of free-form surface mirrors and present solution strategies to take advantage of the information.

Dioptric systems consists of lenses, catoptric systems of mirrors. For the combination of mirrors and lenses the name catadioptric has been established.