Six months ago, Hannes was asked by his friend Joel from the Computer department if he was interested in partnering up with him to write a degree project within AI. The technical part of the Degree Project took place at the company Axis, which Joel worked part time at.
Tell us briefly about your degree project!
- We have investigated AI image recognition on 3D point clouds, generated by a radar. The core of the problem is to be able to look at a certain area within a radar and determine whether there is one or more people in the picture. The application we have tried to carry out is to detect something called piggybacking. We tested it by letting two people try to simulate that they were escaping through a security door when there was only one of them who had access.
Hannes explains that they monitored a hardware which recorded so-called point clouds. Afterwards, they classified a label on every recording which describes the chain of events.
- With help from this, and relatively advanced mathematics, you can teach the network to guess by itself what is happening within the given frames.
What made you say yes to the project?
- My one and only experience with AI was that I knew it was a huge science field and that one of the most central pieces within AI at the moment is image recognition and classification. What made me take on the project was that I have, for a very long time, wanted to better understand how image recognition works and I have always been interested in computer science. I also thought that it would be very interesting to try to apply AI within Supply Chain Management in the future.
It really surprised me how easy it was to go from having zero knowledge, to managing a whole programme.
Hannes describes his limited experience and the process of grasping the science field as one of the biggest challenges.
- I had to prepare myself and learn about AI on my own before the project started since a big challenge was to convince our supervisor that I had enough prior knowledge. But it worked out and I'm happy that our supervisor believed in me.
According to Hannes, it was the process of testing which alternative would give the best accuracy that was the most entertaining during the course of the project.
- The most fun was definitely to sit down and try to find an architecture that works. It felt really good, especially when we saw the significant increases in accuracy. Since we could choose from almost an unlimited amount of alternative methods to increase the results, it was a very fun and challenging experience.
It was a very fun and challenging experience.
How has your education helped you during the project?
- I have tried to contribute with my knowledge in mathematics and statistics, since I have more experience than Joel in those areas. When you study to become an engineer you learn to tackle problems in a logical and structured way. I believe that has been the most valuable and useful. It also helps to have some computer knowledge, which I've learned from programming on my own.
AI-advice from Hannes
As a tip for those students who are considering doing their degree project within AI – but who are afraid to not have enough prior knowledge in the subject – Hannes emphasises that it was easier than expected to learn the subject.
- When it comes to coding, it is very simple to start, especially with user-friendly libraries such as “TensorFlow” and “Keras”, which we used to create our programme. It really surprised me how easy it was to go from having zero knowledge, to managing a whole programme.
- My other recommendation is to be very precise with documenting all your tests and results, and to start your research on time. I would also recommend writing your thesis with someone else – I believe that having a good time with your partner is an important aspect of succeeding with your project.
