What is it about?
Detecting a round object with a distinctive pattern is simple, as long as the sun is not directly shining on the soccer bal, changing its appeareance. By first creating a model of how light reflects from its surface, an infinite number of training images can be created, which makes it possible to train a ball detection algorithm further, making it robust for all lighting conditions.
Featured Image
Photo by Patrick Schneider on Unsplash
Why is it important?
Many computer vision algorithms are trained on datasets that were recorded under good lighting conditions. Once these algorithms have to be applied outside the lab (security cameras, autonomous driving vehicles, robots, etc), it is important to add robustness against the lighting conditions which can be encountered in the real world. With this model any lighting condition can be generated (although in this case only for one object; a soccer ball).
Perspectives
I expected that generating such a shape and reflectance model was the hardest part. Yet, integrating the results into a digital twin of the soccer field in our lab was maybe even harder. Yet, having such a digital twin which allows to generate training images in a controlled way was worth the effort.
Dr. Arnoud Visser
Universiteit van Amsterdam
Read the Original
This page is a summary of: Using Neural Factorization of Shape and Reflectance for Ball Detection, January 2024, Springer Science + Business Media,
DOI: 10.1007/978-3-031-55015-7_11.
You can read the full text:
Contributors
The following have contributed to this page
