Sebastian Rietsch

Advisors
Pjilipp Schlieper (M.Sc.), Prof. Dr. Björn Eskofier

Duration
07/2019 – 02/2020

Abstract
The goal of this thesis is to explore solutions for adapting a drone detector to a nighttime environment. The dataset at hand largely consists of labeled daytime RGB images with only a relatively small portion of labeled nighttime data in the form of near-infrared images, which were captured from one static camera only. Training the drone detector on this data directly results in insufficient nighttime performance.

This highly relates to the research area of domain adaptation, which formally refers to the goal of learning a concept from labeled data in a source domain that performs well on a different but related target domain [4].

Newest research shows promising results, even though most of it is focused on the application of image classification and only little research is directly related to object detection. The content of this thesis will be to first identify promising approaches which can be applied to the object detection problem. Afterwards they will be evaluated on real data and compared to each. The evaluation will be performed via a suitable, preferably un-biased state-of-the-art object detection algorithm and other (application independent) evaluation metrics. The outcome of this thesis will show, to what extent domain adaptation proves viable in a drone detection scenario with very limited target domain data.

References:

1. Sathyamoorthy, Dinesh. “A Review of Security Threats of Unmanned Aerial Vehicles and Mitigation Steps.” The Journal of Defence and Security 6 (October 2, 2015): In press.
2. “Tarsier | World-class drone detection.” [Online]. Available: https://www.tarsier.co.
3. Goodfellow, Ian, Yoshua Bengio, and Aaron Courville. Deep Learning. MIT Press, 2016.
4. Wilson, Garrett, and Diane J. Cook. “A Survey of Unsupervised Deep Domain Adaptation.” ArXiv:1812.02849 [Cs, Stat], Dec. 6, 2018.