Datenbestand vom 12. August 2022
Tel: 0175 / 9263392
Mo - Fr, 9 - 12 Uhr
Fax: 089 / 66060799
aktualisiert am 12. August 2022
978-3-8439-3043-7, Reihe Statistik
Efficient statistical analysis of video and image data
187 Seiten, Dissertation Ludwig-Maximilians-Universität München (2016), Softcover, A5
Due to the fast technological progress of our days, statistics is faced with various methodological challenges. In many different areas, video and image data are collected and cannot be analyzed manually due to the large data volume. The present thesis deals with two complex problems of applied statistics which are motivated related to the global climate change. In both interdisciplinary projects, statistical methods are developed to extract relevant information from video and image data efficiently and with low manual effort.
The first project originates from the area of fisheries ecology:
With the expansion of renewable energies, more and more water power plants are constructed and make fish migration difficult.
Motivated by this fact, in this thesis a system is developed which allows to count and classify fish seen on underwater sonar videos in front of water power plants automatically and in realtime.
With the information about number and species of fish, protection measures can be taken to help the fish to migrate through rivers.
Within the scope of this project, a software was implemented which allows to apply the developed system at a water power plant during operation.
The second project originates from the area of phenology:
The climate change research community is interested in the question if season onset dates change due to the global warming.
To observe season onset dates for a large amount of different locations, webcam images can be used.
In this thesis, a method is developed which allows to automatically extract season onset dates from webcam images.
The usefulness of the developed methods is demonstrated with data from two scientific webcams and three webcams with data publicly available from the internet.
Moreover, by analyzing images from a publicly available webcam database of over 13000 webcams it is shown that the developed methods can be applied completely automatically to large data volumes as well.
All developed methods are implemented in the statistical software package R and publicly available in the R packages sonar and phenofun.