This repository showcases the classfication of contact between a finger and an inert surface, such as a tabletop, in a video stream produced by a depth camera, through a supervised learning approach.
The targeted interaction can be observed in the video from the paper:
A series of notebooks shows how to:
These produce the model that was used for the experiment.
- install required python package from requirements.txt:
conda install --file requirements.txt
- install additional dependencies:
brew install pcl
pip install python-pcl
conda install -c open3d-admin open3d
- compile cython extention
python helper/deproject/setup.py build_ext --inplace
- install the dataset from the at the root of the repository in the folder called
dataset.
This repository includes the dataset on which the model was trained and tested. It is accessible in the release.
If you want to learn more about potential applications, please refer to the associated paper Gesture Typing on Virtual Tabletop.
@inproceedings{Loriette:2017:GTV:3132272.3135074,
author = {Loriette, Antoine and Murray-Smith, Roderick and Stein, Sebastian and Williamson, John},
title = {Gesture Typing on Virtual Tabletop: Effect of Input Dimensions on Performance},
booktitle = {Proceedings of the 2017 ACM International Conference on Interactive Surfaces and Spaces},
series = {ISS '17},
year = {2017},
url = {http://doi.acm.org/10.1145/3132272.3135074},
}