Nachricht

Am Mittwoch, den 5. Dezember 2018 hält Herr Mawe Sprenger um 16:30 Uhr im Raum P 1.4.17 einen Zwischenvortrag über seine Masterarbeit mit dem Titel:

Convolutional Neural Network for Depth and Odometry Estimation from Monocular Video

Abstract:

Dense 3D maps are becoming increasingly important for navigation tasks of autonomous mobile robots. Although various depth sensors can already be used for dense 3D mapping, ongoing research is conducted in depth estimation with cheap off-the-shelf commodity cameras. In this thesis, a system for the simultaneous estimation of dense scene depth and camera ego-motion from monocular image streams will be developed and implemented. For this purpose, different deep neural network architectures for depth and odometry estimation are analyzed and evaluated. Subsequently, the best performing networks from both fields are selected and combined into a single end-to-end approach. The developed system will be trained with different levels of supervision and tested against existing methods for depth and odometry estimation. Using the estimated data, 3D maps will be created with the latest available 3D SLAM software and a conclusion will be drawn regarding the system's applicability for 3D mapping.