This software allows to record hand gesture Emg and orientation signals using a surface Emg device. Compatible devices are Myo Armband and gForce Pro.
The data collected in this interface is fully compatible with Manual Segmentation.
- Check the list of data collectors in the excel file configs/list of collectors.xlsx.
- Check the list of devices in the excel file configs/list of devices.xlsx. Go to the repository root folder and just execute the script ejecutar.m in matlab.
Using the Myo Armband has an easier installation procedure. Check the MyoMex repository for the tutorial installation. We have included in this repository a copy of MyoMex.
The gForce_mex function must be compiled and added to the path. If you receive an error about it, check the gForce_interface project for help.
Be sure to have correctly install gForce Mex. Check the Installation Guide for the complete tutorial.
We recommend any of the following 2 options to include gForce:
- copy and paste in this folder the gForce package (i.e. /export_mexMatlab/), or
- add the gForce directory to the Matlab path.
The compatible devices have a couple of difference that were taken into account in our system to make the data recorded compatible between devices.
- Emg sampling rate: Myo Armband samples at 200Hz, whereas gForce Pro, was set at 500Hz (although it can up to 1kHz). The sampling rate is specified in a metaparameter in the userData.mat file.
- Emg range: Both devices have Emg data in the range [-1 1].
- Device prediction: Only when Myo Armband chosen it is included the prediction field. Refer to gForce SDK issues for more info.
- Accelerometer and gyroscope data: Only when Myo Armband chosen it is included these fields. Empty in the case of gForce Pro.
In the following figure there is the mapping of the channels of each device, the yellow block marks the "logo electrode". For compatibility, we rotate the Emg matrix of the gForce Pro to match Myo Armband channel distribution.
Data recorded for every user is stored in folder ./data/ in *.mat format. The software creates a new folder for each user with the given name removing spaces and special characters.
-
The gForce library is still in beta. There are some issues with the SDK that may difficult data acquisition. Please, be aware of the following issues:
-
In the case of a Run Time Error in the C++ MEX function, Matlab will crash inmediately and close itself. The device must be restarted manually.
-
Battery level sometimes returns 0%. This is the response from the device (the SDK not the C++ MEX function). It may not correspond to a completely empty battery, as in some tests afterwards the command returned a value (e.g. 77%).
-
Emg data is encoded in 8 or 12 bits. So, it should be centered at 128 or 2048, but it is not. The reference is slightly moved down. At 8 bits it is centered in 118 and at 12 bits, in 2000.
-
Gesture predictions are currently unavailable due to empty responses from the device.
We included in this repository the MyoMex connector. Special thanks to @mark-toma.
After finishing the recordings in this interface, to carry on with the manual segmentation procedure, do the following:
- Copy the data of all users (i.e. the content of this ./data/ folder except the README.md file) to the /manualSegmentation/data/ folder in Manual Segmentation repository.
- Run the script /manualSegmentation/ejecutar.m in manual segmentation repository.
If you use this Data Acquisition System, please refer and cite the following paper:
Changelog:
- v1.7 | June 21st, 2024 | Inclues GUI to survey and optimization of the data collection process.
- v1.6 | June 26th, 2023 | Inclues GUI to gesture set selection.
- v1.5.1 Updated recolectors and list of devices.
- v1.5 Version of dataset G11.