Michael Vössing, Niklas Kühl

Background:

For car insurers, the assessment of car damages is one of the most important processes. Since the manual detection and assessment of damage takes a lot of time and is also a monotonous and error-prone task, automatic image processing is recommended. HUK-COBURG is striving for such a solution with the help of state-of-the-art algorithms in the field of machine learning and computer vision. A possible solution to the problem consists of two tasks: Firstly, car part recognition is used to distinguish between the different car parts (see PDF). Then the individual parts are analysed to determine whether they are damaged or not (see PDF). Although the machine learning algorithms have achieved good results in many domains, some studies show that collaboration with the domain expert can further improve these results. By so-called "human-computer collaboration" not only the machine learning predictions are considered, but they could be supplemented or corrected through domain experts if necessary. In the presented use case, this idea can be applied to multiple aspects: The expert can correct the incorrectly recognized parts or the incorrectly analysed damage. Further, the system could interactively ask the expert to take additional photos or guide him to take better photos.

Research Goal:

In this thesis, we want to study how human-computer collaboration and interaction should be integrated into the damage recognition system. The complexity and diversity of possible car damages in the car insurance sector, as well as the "car damage analyser" offered by HUK-COBURG, provide a suitable environment for this research.

Working on this thesis you will:

• - apply your theoretical knowledge to a practical use case.
• - further develop and evaluate existing computer vision models.
• - become an expert in deep learning.
• - work directly with domain experts to improve the usability of the system.

We look forward to receiving your application because you...:

• - have solid programming skills in Python (e.g., pandas, scikit-learn)
• - have experience with state-of-the-art computer vision and deep learning frameworks (e.g., TensorFlow, YOLO,OpenCV)—or are willing to learn quickly.
• - have an understanding of Design Science Research—or are willing to learn quickly.
• - want to gain first-hand experience solving real-world problems

Details:

• Start: as soon as possible
• Duration: 6 month
• Salary: Yes
• Location: HUK-COBURG, Coburg or Remote

PDF