08/2021 – 01/2022
Behavioural monitoring of captive animals is an important tool for biologists to better understand these animals, their physical and psychological health and to analyse observed behavioural changes and the possible reasons behind them. With this information, the care for the animals and the environment they are living in can be improved to support animal welfare. However, the current monitoring process is very labour-intensive, because the most common practice is that the observer watches the animals repeatedly on-site over a long period of time and documents their behaviour manually with pen and paper [1, 2]. Automating this process would significantly reduce the effort to acquire data and would make it easier for biologists to perform long-term studies.
In principle, the methods used to automatically record and analyse the behaviour of individual animals can be categorized in two classes: invasive methods which attach devices like GPS transmitters to the animals to track them, and non-invasive methods which monitor the animals from afar, for example with cameras. Non-invasive methods are preferable as they cannot harm the animal in any way and do not change the animal’s behaviour as it could be the case with an attached device [2, 3]. Recording videos of the animals with the help of cameras is a simple setup and it is also easy to detect animals in these videos with the help of object detection algorithms which provide good results and are commonly available since recent years . But the main challenge for a fully automated behavioural analysis is the identification of individual animals in the videos. This challenge can be tackled by using re-identification (Re-ID) methods. A Re-ID approach compares the recorded video or image of an individual (query) with a list of known identities (gallery) and returns an ordered list of identities that have the most similarities to the recorded individual. Ideally, this also works if the query and the gallery video or image were captured from different cameras, under different light conditions, with occlusions or at a different time .
Re-ID methods are mostly studied in the context of person-based surveillance systems and there is little research on using it for animals [5, 6]. Most existing approaches are image-based or do only work with species specific features [7, 8]. Some of the current state-of-the-art image-based person Re-ID methods detect human body parts to improve performance but these approaches are not easily applicable to animals as they have different postures than humans [6, 9]. On the other hand, some video-based methods use the spatial and temporal information from a video sequence to reach a similar or higher accuracy without any part-based detection [9, 10]. Therefore, this thesis will create a video-based Re-ID datasets of captive polar bears and will evaluate if existing video-based person Re-ID methods can be trained to identify individual polar bears from the dataset.
 Jason Watters, Susan Margulis, Sylvia Atsalis, (2009), Behavioral Monitoring in Zoos and Aquariums: A Tool for Guiding Husbandry and Directing Research, Zoo biology, Volume 28, Pages 35-48
 Marian Stamp Dawkins, (2007), Observing Animal Behaviour – Design and Analysis of Quantitative Data, Oxford Biology, Oxford University Press  Jeffrey Rushen, Anne Marie de Passille, N. Chapinal, (2012), Automated monitoring of behavioural-based animal welfare indicators, Animal Welfare, Volume 21, Pages 339-350
 Stefan Schneider, Graham W. Taylor, Stefan C. Kremer, (2018), Deep Learning Object Detection Methods for Ecological Camera Trap Data, 15th Conference on Computer and Robot Vision (CRV), Pages 321-328
 Ehsan Yaghoubi, Aruna Kumar, Hugo Proenca, (2021), SSS-PR: A short survey of surveys in person re-identification, Pattern Recognition Letters, Volume 143, Pages 50-57
 Prashanth Ravoor, Sudarshan Tsb, (2020), Deep Learning Methods for Multi-Species Animal Re-identification and Tracking a Survey, Computer Science Review, Volume 38, Page 100289
 Stefan Schneider, Graham W. Taylor, Stefan C. Kremer, (2020), Similarity Learning Networks for Animal Individual Re-Identification – Beyond the Capabilities of a Human Observer, 2020 IEEE Winter Applications of Computer Vision Workshops (WACVW), Pages 44-52
 Shuyuan Li, Jianguo Li, Hanlin Tang, Rui Qian, Weiyao Lin, (2020), ATRW: A Benchmark for Amur Tiger Re-identification in the Wild, Proceedings of the 28th ACM International Conference on Multimedia, Pages 2590-2598
 Mang Ye, Jianbing Shen, Gaojie Lin, Tao Xiang, Ling Shao, Steven C. H. Hoi, (2021), Deep Learning for Person Re-identification: A Survey and Outlook, arXiv:2001.04193
 Jianing Li, Jingdong Wang, Qi Tian, Wen Gao, Shiliang Zhang, (2020), Global-Local Temporal Representations For Video Person Re-Identification, IEEE Transactions on Image Processing, Volume 29, Pages 4461-4473