Mosaicing with Pan-Tilt-Zoom Cameras

30 de outubro de 2020

PTZ cameras become popular over the last years in activities such as traffic monitoring, sports analysis and surveillance systems due to its ability to cover a wide range of views at a high definition over multiple zoom levels. Having an overall perception of a scene become significantly important in such activities and subsequent applications, thus the need for its panoramic background representation. This dissertation work is focused on the PTZ panoramic image generation process over multiple zoom levels. The work involves studying different camera calibration algorithms, whose results are then used as input of the panoramic transformation step. Performance metrics are established to assess pixel color errors and perceived quality of the given panoramic transformations in the cube and cylinder based representations. It is further proposed, a variance minimization approach aiming at correcting calibration errors, namely the focal length parameters and consequently, improve the final mosaics. The last part of this work consists of studying and applying different image blending techniques. This step compensates for exposure differences and reduces generated artifacts in the final images. An image blending method is implemented where the final mosaic is reconstructed from its combined gradient. The main mathematical tool used in the blending method is the Poisson partial differential equation with Dirichlet boundary conditions. The proposed methods allowed to obtain promising results on panoramic image mosaicing.

Autores da comunidade :

• 

  R

  Rodrigo Franco Serra

  ist181245

Orientadores desta instituição:

• 

  J

  José António Da Cruz Pinto Gaspar

  ist13495

electrical-engineering-electronic-engineering-information-engineering - Engenharia Eletrotécnica, Eletrónica e Informática

eng - Inglês

Instituto Superior Técnico