Status: available Supervisor: Daniel Helm, Martin Kampel
Start: as soon as possible
Automatic video content analysis is needed in many different domains such as historical film preservation or film archiving. Due to the enormous quantity of available visual media data, automatic approaches shall be used in order to find specific content in videos or create new visual representations of specific domains. A fundamental part for analyzing large film archives is the detection and classification of camera movements such as PAN, TILT or TRACK within a given image-sequence.
The goal of this practical work is as follows:
For establishing an automatic video analysis platform, a “Camera Operations Detection module”, shall be developed. This module shall be able to detect and classify camera movements such as PAN, TILT, TRACK or NO MOVEMENT as well as ZOOM IN/OUT and maybe combinations of those in given input videos. Moreover, a mechanism shall be developed in order to categories a video shot as HIGH-ANGLE or LOW-ANGLE recording. Deep learning-based as well as traditional-computer vision-based approaches shall be explored and evaluated and a dataset should be created (extended) in order to form a fundamental evaluation base.
- Literature Review – getting to know the algorithms
- Implenetation of state-of-the-art approaches (Deep-learning based optical flow vs. camera pose estimation of monocular camera systems)
- Filtering mechanisms for moving/non-moving objects in a scene
- Implementation of an usable Camera Movement Classification module
- Evaluation (qualitative vs. quantitative)
- Written Report and final presentation
Basic knowledge in computer vision Experience in Python. Interest in Deep Learning (Tensorflow, Keras, PyTorch) and Machine Learning
This work is part of the project “Visual History of the Holocaust”.
The practical course is part of an ongoing research project. A “Forschungsbeihilfe” is available for the selected student.