Automatic Panorama Stitching
Theoretical (Analytical):
Practical (Implementation):
Literature Work:
Overview
Soccer tracking is often performed using multi-camera systems. Multi-Camera Panorama stitching is a common method for generating high-resolution video, which serves as the basis for the tracking. Additionally, projection of the stitched video into a single image plane is often accompanied by image distortion, which might cause problems and reduce the tracking quality.
Problem Statement
- Merging of low-overlap videos is often done manually or semi-automatically.
- Distortion degrades the quality of further computer vision methods
- It is desirable to identify distortion-minimizing projection methods
Tasks
- Implement automatic alignment and calibration of camera parameters from low-overlap videos
- (Optional) automatic estimation of lens distortion parameters
- Evaluation of different projection methods for the mapping between camera views and the output video with a focus on minimizing distortion of the input videos
- Evaluation of the application potential of the researched methods as well as potential future work towards a completely automatic stitching and reprojection method
Requirements
- Good Python and OpenCV skills
- Basic knowledge in computer vision fundamentals
- (Optional): Basic knowledge about deep learning models
Scope/Duration/Start
- Scope: Bachelor/Master
- 3 Month Project, 3/6 Month Thesis
- Start: immediately