Event-based Unsupervised Learning of Epipolar Geometry
The stereo matching problem is a well-known bottleneck in stereo-vision. Event-based stereo matching approaches make explicit use of precise interocular temporal coincidences. This computationally expensive process can be greatly improved by imposing constraints from the geometry of the stereo setup. However, obtaining this geometry information requires precise camera calibration procedures, which are time consuming and prone to errors. In this discussion group we will investigate the use of event-based learning methods for estimating the stereo setup geometry without requiring prior knowledge about sensors' pose.
|Wed, 25.04.2018||22:00 - 23:00||Lobby|