Automatic damage detection
The roads in Germany are getting crowded. In the last ten years, the number of new car registrations has risen by over 30 percent. As the number of road users increases, so does the number of claims reported to insurance companies. Trained employees check the data and images of each case and decide whether the use of an expert is necessary. The expert opinions cost several thousand euros and represent a not insignificant part of the turnover of an insurance company.
Automated image processing has the potential to simplify this task considerably. In recent years, this field of research has made great progress. The reason for this is the use of so-called deep learning systems, which learn abstract representations on the basis of data in order to solve a wide variety of tasks. At L3S and the Institute of Information Processing, Prof. Dr. Bodo Rosenhahn and his team are researching new methods for automated image analysis. They are to be used to identify objects and their properties in images – including damage to car components.
Together with the Startup ClaimBuddy, a spin-off of the Leibniz University of Hanover in the field of digital claims management, Prof. Rosenhahn and Frederik Schubert are working on a funding project to meet the requirements for automated claims recognition. This includes building a data set to train a deep learning system that can support insurance companies in the analysis and evaluation of claims. For insurers, this offers the possibility of a cost-saving and time-efficient settlement, especially of small claims.
Caption: Scratch detection
Source: Car Damage Detective Dataset
Frederik Schubert is a research assistant at L3S and the Institute for Information Processing. He deals with reinforcement learning and automatic scene analysis.
Bodo Rosenhahn is director of the L3S and heads the Institute for Information Processing. His research is in the fields of Computer Vision and Big Data.