Axis enables a smarter and safer world by creating network solutions that provide insights for improving security and new ways of doing business. As the industry leader in network video, Axis offers products and services for video surveillance and analytics, access control, and audio systems. Axis has more than 3,000 dedicated employees in over 50 countries and collaborates with partners worldwide to deliver customer solutions.
Category
Master of Science in applied mathematics; computer science or similar
Scope
2 students completing 30 credits (20 weeks) each.
Educational Background
Master of Science in applied mathematics, computer science or similar. Understanding of how a Linux system works and experience from courses in machine learning is required. Programming experience in, Python and C/C++ is meriting.
Background
Deep learning is a type of machine learning involving neural networks in a way which is loosely inspired by the way the brain processes information. This branch of machine learning has in the past few years had an amazing success, creating a huge boom which is currently transforming the way entire businesses work.
In the area of video surveillance, deep learning is an enabler for autonomous, or semi-autonomous surveillance systems, relying less and less on the constant monitoring of human operators. Typical algorithms that now predominantly start relying on deep learning are object classification, detection and tracking, and various forms of identification or re-identification algorithms.
However, one of the biggest challenges when working with machine learning is that such algorithms are extremely data hungry, often requiring millions of annotated training images to be used for so-called supervised learning. The task of creating such datasets and annotating them requires an enormous effort that often cannot be automated.
Goal
In addition to synthetic data being completely free from data integrity related issues, it also makes it possible to get the annotation almost for free, since all of the information about the scene is already known by the computer engine generating the scene. This opens up for the possibility of using more advanced algorithms working in the time-domain rather than just analyzing still images.
There are at least three possible main goals for this thesis:
The first goal is to find out how good results one can get by generating large amounts of artificial training data and training a machine learning algorithm using this data. The second goal is to enhance the training data further using Deep Learning methods. The third goal is to analyze temporal algorithms which typically benefit from even larger variation in the training data than is possible today by using real-world video only to train on.
OK, I am interested! What do I do now?
You are valuable to us – how nice that you are interested in one of our proposals! There are a few things for you to keep in mind when applying.
We look forward to hearing from you!
25-03-2024
Ange nedan vart du önskar arbeta och glöm inte bort att ange din e-postadress!