Collect gestures data package

Description

The ROS package record data from 2 IMU MetaMotionR+ from Metawear (Hardware datasheet, Software documentation). Each data is saved in a separated file. These data would later be used to train a classification model. This program output uniformised data, once calibrated. It also include the node ros_metawear_simple which publishes the data of the sensor fusion from the IMU. Finally, in the notebook directory you can find the notebook used to process the data and train the model.

Installation

• This program run in python 2.7.
• Clone this repository inside your desired catkin_ws/src then build it.

Dependencies

• Python packages: Numpy, PyKDL, XLib.
• pyxhook is already provided in the folder.
• ROS: Already included messages.
• For the streaming node, one need also: mbientlab.metawear package

Utilisation

Command to launch the collecting node: rosrun collect_gesture_data collect_data.py _address_imu_r:=E5:4D:16:18:CD:90 _address_imu_l:=F1:38:44:7A:F9:99 _gesture:=aggregate _calibrate:=False _session:=L1

• _gesture is the name/label of the performed gesture.
• _calibrate is a boolean to say if one want to calibrate the sensor. It means that it align the axis with X-front, Y-left and Z-top. Must be down at the beginning of each session.
• _session is a string giving the name of the session. If _calibration is true, calibrate then save the parameters with this name. Otherwise take the saved parameters under this name.

Command to launch the streaming node: rosrun collect_gesture_data ros_metawear_simple.py _address:=E5:4D:16:18:CD:90

• _address is the MAC address of the IMU.

It subscribes to:

• /metawear_ros_E5_4D_16_18_CD_90/rotation: gives orientation of IMU (quaternion).
• /metawear_ros_E5_4D_16_18_CD_90/accel: gives acceleration of IMU (g).
• /metawear_ros_E5_4D_16_18_CD_90/gyro: gives angular speed of IMU (rad/s).

And that for both IMU.

Instruction for use

• (Optionnal) Calibrate sensor.
• (Optionnal) Change the directory where data are saved.
• Double click on your mouse to start/stop the recording of the gesture.
• The data from the different topics are synchronized based on their time stamp.
• The out put is a csv file with linear acceleration ( no gravity), rotation in quaternion and the angular velocity around the corrected axis.