High performance bees tracking solution

How to use the solution

This project contains 2 targets:

• example.cpp: an simple example program which run the solution
• benchmarks.cpp: target used to run benchmarks and obtain metrics on performance

The example input data for both target can be downloaded and extracted using the dedicated script:

cd data
bash download_data.sh

Please contact me if you can't download the file

How to compile and use the example target

mkdir build && cd build
cmake ..
make -j example
./example

Without any debug information, the example program generates a metadata file which contains all the parameters used in the run, as well as the output statistics

How to compile and use the benchmark target

Please take care that ENABLE_PERFORMANCE_LOG in implementation/Inc/Parameters.hpp is set to true at this point.

mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make -j benchmarks
./benchmarks

The benchmarking process may take a few minutes. Then the performance metrics are generated in ./eval The graphsGenerator.py python script can be used to generate plots from these metrics.

Debug levels

Debug options can be modified by changing DEBUG_FLAGS in implementation/Inc/Parameters.hpp The following options are available:

• logBees: log bees in a csv file
• logTrajectories: log bee trajectories in a csv file
• generateOutputFrames: draw circle around the detected bees on each of the input frames
• generateBeeImages: generate small jpeg image for each recognized bee. After manual labelling, this can be used to train a machine learning algorithm, which may be used to improve accuracy.
• generateIntermediateImages: generate the intermediate states (T0, T1...) of the image processing algorithm

Generate Doxygen documentation

A Doxygen documentation can be generated using the following commands:

cmake ..
make doxygen

The output will be generated in documentation/doxygen/doc

Bibliography

The solution implemented in this project is mainly inspired from the following papers:

• Baptiste Magnier, Ga ̈etan Ekszterowicz, Joseph Laurent, Matthias Rival, Fran ̧cois Pfister. Bee hive traffic monitoring by tracking bee flight paths. 13th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications, January 27-29, 2018, in Funchal, Madeira, Portugal, pp.563-571, 2018, 10.5220/0006628205630571. hal-01940300 https://hal.science/hal-01940300/document
• Baptiste Magnier, Eliyahou Gabbay, Faysal Bougamale, Behrang Moradi, François Pfister, et al.. Multiple honey bees tracking and trajectory modeling. Multimodal Sensing and Artificial Intelligence: Technologies and Applications, Jun 2019, Munich, France. pp.29, ⟨10.1117/12.2526120⟩. ⟨hal-02187412⟩

This project has started as a freestyle exercise during the "Efficient parallel C++" lab course at Karlsruher Institut für Technologie. I would like to thank in particular the supervisors of this course.