The second is probability statistics, because the main goal of computer vision research is to make computers have the ability to understand natural scenes through cameras. Dealing with inferences in real life requires knowledge of probability and statistics.
There are many other aspects of mathematics used in computer vision research, such as discrete mathematics, graph theory, differential geometry, Riemannian geometry, Lie groups and Lie algebras, manifold learning, tensor analysis, principal component analysis, nonlinear optimization and so on.
When doing computer vision research, you don't need to learn all these basic knowledge before doing research. Even if you master all these mathematical knowledge, you may not be able to solve the problems in your research with these mathematical knowledge.
Personal opinion: It is more practical to look for mathematical tools with research questions than to look for problems after mastering mathematical knowledge. Unless you majored in mathematics at first, you'd better be problem-oriented in your research, or you'll learn a lot of basic knowledge of mathematics, and most of it will be useless in the end.
In short, it is good to learn whatever mathematical knowledge is used in the research. There is no need to learn all the relevant content. After all, scientific research is not to cope with the math exam.
If you are interested in mathematics while doing computer vision research, you can pay attention to the latest scientific research progress in mathematics and see what new theories and algorithms appear and whether they can be used in your research direction. That's enough. Solving old problems with new methods is also an effective innovative means.
Finally, it needs to be emphasized that it is absolutely unnecessary to bury yourself in the pile of math books from the beginning when doing research on computer vision.