Skip to content

Protocols

Jump to bottom
nspi edited this page May 3, 2020 · 4 revisions

2020/05/03:

Duration: 10am - 11am

Participants: Samad, Mohammad, Nico

Reports:

Discussion:

  • Samad and Mohammad see tracking of multiple persons as a possible future problem.

  • Additionally, once a person has fallen, openpifpaf does not detect them anymore.

Todo:

  • Combine both features (rotation energy, generalized force) into a single algorithm.

  • Download the video data from the UP-Fall Detection Dataset and re-organize it such that there is one folder for each activity containing the videos from all subjects.

  • Develop a version of the algorithm without any GUI elements that runs similar to a "batch" process on all videos within a folder and stores the data.

  • Develop a script that plots the computed data for each folder/activity in a single plot

Next meeting: Date will be discussed in Telegram

2020/05/01:

Duration: 5pm - 6pm

Participants: Samad, Mohammad, Nico

Reports:

  • The algorithm was restructured, making data exchange between parallel processes more effective

  • Nico will ask Prof Deserno to join us in week 20 and writes another email regarding the dataset

Next meeting: 2020/05/03

2020/04/27:

Duration: 1pm - 2pm

Participants: Samad, Mohammad, Nico

Reports:

Discussion:

  • Nico will write an e-mail to the authors if they can provide the video data. If they do not reply, we will make use of the UP-Fall Detection Dataset

2020/04/22:

Duration: 1pm - 2pm

Participants: Samad, Mohammad, Nico

Reports:

  • Openpifpaf is working nicely with and without CUDA support

  • An interface was developed that stores the data in .CSV files

Discussion:

  • We need to run openpifpaf in a thread to add the algorithm for fall detection in a concurrent thread

  • Subsequently, we will decide if Samad and Mohammad will work on the same algorithm for fall detection or develop different algorithms in parallel.

Next meeting: 2020/04/24 or 2020/04/25 depending on how easily multithreading can be implemented

2020/04/20:

Duration: 1pm - 2pm

Participants: Mohammad, Nico

Discussion:

  • We discuss a first algorithm. We decide to use OpenPifPaf for pose estimation and develop an interface to acquire the data

Next meeting: 2020/04/22 1pm

2020/04/17:

Duration: 8pm - 9pm

Participants: Mohammad, Samad, Nico

Reports:

  • Mohammad took a look at implementations with OpenPose not running but OpenPifPaf running smoothly on his laptop

  • Samad took a look at research papers and summarized them (Google Docs)

Discussion:

  • We discuss the general structure of the algorithm. From a meta perspective it will be something like:

Todo:

  • Fill the (Google Docs) with relevant papers and algorithms

  • Get to run favorite algorithm and demonstrate it in next meeting

Next meeting: 2020/04/20 1pm

Clone this wiki locally