A Fast and Accurate Object Detection Algorithm on Humanoid Marathon Robot (Hurocup Marathon)

This is a repository that implement color based region segmentation and marker detection using Convolutional Neural Network(CNN) that is used on FIRA Hurocup Marathon Game. This method able to segment out the Marker and detect its respective direction whether left, right or forward that indicate the continuous action that need to follow by robot.

Acknowledgement

Thank for the paper "A Fast and Accurate Object Detection Algorithm on Humanoid Marathon Robot" which allow me to work on this repository. I would suggest you all to have a look on the paper to understand more about the effort and work.

Requirements

Python Version: 3.6.10
Install the require dependency with this command:

$ pip install -r requirements.txt

$ pip3 install -r requirements.txt

Download the data from Kaggle and organize into:

I would suggest the train.py script run in the GPU environment as its tend to save your time, or you can try to run the training at the Google Colab which provides free GPU.

Usage

To run the classification scripts:

$ python classify.py --model arrow.model --labelbin lb.pickle --image examples/sample_up.png

To run the training scripts:

$ python train.py --dataset dataset --model arrow.model --labelbin lb.pickle

Work

The CNN model architecture is created and located in model/marathonnet.py

Note

I didn't implement the cross-validation strategy. Don't get me wrong, cross-validation are important to prevent model overfitting. Anyone are invited to submit their pull request to further improve the work.

Reference

Jamzuri, E. R., Mandala, H., & Baltes, J. (2020). A fast and accurate object detection algorithm on humanoid marathon robot. Indonesian Journal of Electrical Engineering and Informatics, 8(1), 204-214. https://doi.org/10.11591/ijeei.v8i1.1960