Google Search
Jun.-Prof. Dr. Matthias Hirth
Head of Group
Secretary
Marina Bondarev
+49 3677-69 2890 | Fax: +49 3677-69 2888
marina.bondarev@tu-ilmenau.de
Visitor address
Gustav-Kirchhoff-Straße 1
Kirchhoffbau Room K 3007 (2nd upper floor)
98693 Ilmenau
Postal address
Technische Universität Ilmenau
Fakultät für Elektrotechnik und Informationstechnik
Institut für Medientechnik
Postfach 10 05 65
98684 Ilmenau
Funded by: ScaleHub
Duration: 2020 - 2022
Contact: Jun.-Prof. Dr. Matthias Hirth, Edwin Gamboa, M. Eng., Dan Dubiner (ScaleHub)
Artificial intelligence (AI) is one of the leading business drivers today. The recent advances in algorithm research and hardware development enable novel applications that have been unthinkable a few years ago and dramatically change existing industries. Concerning the traditional German industry sectors, the digitalization and application of AI in the automotive field might be one of the most prominent examples. German car manufacturers were well known for their innovations and technical quality for decades, but now new global players like Tesla, Google, or other companies from the IT sector emerge in this field. In contrast to traditional manufactures, the strength of these companies lies in developing software and AI solutions that enable convenience features like (semi-) autonomous driving.
The critical requirement for developing AI solutions for autonomous driving, and AI solutions in general, are massive and very well curated training datasets. In some cases, such datasets are already available, e.g., if they can be derived from administrative data or sensor readings. Sometimes, it is also possible for AI solutions to automatically generate datasets and reuse them for training and improvement. This approach was used, for example, by Google’s AlphaGo Zero. However, in most cases, and especially for applications in the audio-visual domain, humans still need to generate the datasets manually. Google uses mechanisms like reCaptcha to utilize the working power of hundreds of thousands of internet users for image categorization and segmentations for free. Most other companies, like Amazon, focus on manual annotation with inhouse employees or crowdsourcing-bases approaches to provide similar services. However, these often come at high personnel costs.
In this context, the goals of this project are two-fold. The first goal is to improve existing crowdsourcing-bases approaches for the segmentation of images, videos, and volumetric data with respect to their costs and accuracy. The second goal is to develop AI solutions to support the workers during the segmentation process.