Depth estimation in video streams based on an object database

OData support
Dr. Kovács Gábor
Department of Telecommunications and Media Informatics

This thesis discusses depth perception, in which my goal was to determine the distance of detectable objects from the camera in various video records. To determine this value based on one picture is far from an unequivocal task, thus methods used in practice usually involve two or several cameras.

After gaining deeper insight into the subject via studying, my focus turned to the creation of a system which is able to estimate the depth of different types of objects of video records, based on a previously created database.

During the building of the database, my goal was to store information which are directly or indirectly related to the distance of the object. This includes the circumscribed rectangle or contour of the object, or the lenght of the displacement vector.

The development of the distance estimating system was accomplished in a way that it is capable of managing the content of the database by different aspects and configurations. In order to successfully reproduce the measurements, the information required for machine learning were stored in the object database, as well as the test data used in the measurements.

Following the implementation of the system and setup of the test environment, the distance estimating mechanisms used in the system were evaluated based on the test results.  


Please sign in to download the files of this thesis.