||A bee hive is a suitable model to study disease transmission in human society. Much like humans, bees interact in ways that allow for disease transmission, and some bees have more interactions with more peers than others. Researchers at the department of Ecology, Swedish University of Agricultural Sciences, are developing methods to quantify this type of interaction and derive interaction networks encompassing all bees in a hive, individually identified. Video is recorded of bees in an observation hive, after tagging each bee with a unique identifier. These videos need to be analysed fully automatically to identify each interaction (which bees interact, when, and in what way).
Bee tracking software needs to be implemented. It is not clear which tracking algorithm will work best for this particular problem. Challenges include the large number of bees to be tracked (600-900), the large data rate in the video stream, and the physical closeness of bees in the hive. Ideally, individual bees will be followed, such that each track segment can later be used to identify the bee by reading its tag (rather than reading the tags in each video frame). When bees walk over each other, it is expected that the algorithm will not be able to track the occluded bee. The algorithm will have to recognize this and end the track, and then start a new track when the bee is visible again. By reading the tag on the bee, it will later be possible to link these two track segments.
The algorithms will be tested on existing videos recorded last spring. It is expected that a working algorithm will be available by mid May, when new videos can be recorded. The new videos will differ from the currently available ones in that we will try to improve the illumination; also the tags used on the bees will probably be different. The algorithm will have to be adapted to work well with the new imaging setup.
All algorithms will be written in either MATLAB or C/C++. MATLAB is ideal for rapid prototyping, but, if time permits, some critical portions could be rewritten in C or C++ for better performance. The work will be done at the Centre for Image Analysis.
Depending on progress, the tasks for this thesis can be extended by, for example, rewriting portions of the algorithm to use the GPU for increased computing power. Another task within the larger project that could be tackled is recognizing the moments of interaction, and identifying the type of interaction.