First are explained data of the experimentations. In other words, the correspondence between the name of the files and what has been measured is shown. Main results of the experimentation follows.
Experimentation
Python files are for the experiment : reception.py, processing.py and kalman.py. They are from the Kalman-Filter repository (https://github.com/JulienMellet/Kalman-Filter) ; test.py uses the Plotter to draw patterns and save measurements data into a text file ; data_processing.py draws graphs from data collected.
Two experiment have been done to know how change the accuracy of tracking with LH positions.
- Experiment with LH looking in the same direction.
data_Kalman_* and data_Madgwick_* are two types of data taken, as showing on the picture, along a line. Filter of LH data use a mean of 4 last data and IMU position is reset at 30 Hz.
- Experiment with LH looking each other.
- Along a line at important localization (Data Data_2 Data_3 on the scheme)
- Oscillations at different amplitudes
- Round-trip movement of 10 cm amplitude data_2_Kalman_Sin_1
- Round-trip movement of 5 cm amplitude data_2_Kalman_Sin_2
- Round-trip movement of 20 cm amplitude data_2_Kalman_Sin_3
- Oscillations at different amplitudes
- Low pass filter on optical measurement
- Mean of 4 last values are on standard measurement
- 10% of new data data_2_filter01_Sin_2
- 50% of new data data_2_filter05_Sin_2
- IMU refreshment
- Standard refresh data frequency is 30 Hz
- 2 Hz refreshment with static tracker data_2_IMU05s_static
- 2 Hz refreshment with tracker in oscillations data_2_IMU05s_Sin_2
Main results
Graphs and data processing are available on these following google sheets. There
- Lighthouses looking in the same direction https://docs.google.com/spreadsheets/d/1U9xUP7-oBSLRfHlx0UvoTQzaMAvFugudtHTGALtR2J4/edit?usp=sharing
On the left is represented a calculated deplacement of 1cm along the line. To the right is the claculated 2D displacement of the tracker.
Accuracy is not constant and the displacement of 1 cm is not respected. But the line in the space is almost correct.
- LightHouses looking eachother https://docs.google.com/spreadsheets/d/1cFtR_RXirnSVjp703GaSUJfKXmqBP_hrrPQl3wcJRR8/edit?usp=sharing
Like previous graphs, on the left is represented a calculated deplacement of 1cm along the line. To the right is the claculated 2D displacement of the tracker.
The 1 cm displacement is more respected in this case.
Here are presented measurements on an alternative displacement. And find also a representation of data before and after fusion.
Sensor fusion is not useful because standard deviation of optical measurement seems to be preponderant on standard deviation of the IMU. 0,1 filter on optical measurement is smoother than others. Then Madgwick or Kalman filter have comparable results.
- Discussion
Accuracy of 3D tracking seems to depend of the 3D position of the tracker. Accuracy of the optical postitionning system seems to have to be improved to see effects of the data fusion with the IMU.