|
- Date: 17.06.2013
- Time: 11:00am
- Location: Room 02.13.010, FMI-Building, Garching, Boltzmannstr. 3
|
Contributions to Stereo Vision
Abstract:
This work presents camera-based methods for dense scene reconstruction. In the context of binocular stereo vision, we introduce a novel algorithm for the efficient computation of dense disparity maps and a precise stereo method that is based on simulated random walks. Further, we present a novel method for the robust and efficient reconstruction from multiple views using a probabilistic formulation. In practice, we use these methods to realize various automotive parking assistance functionalities including an automatic parking slot detection, a collision detector for the pivoting ranges of the doors, and novel image based rendering techniques for visualization. The influence of reconstruction quality on application reliability is showcased with extensive experiments.
