EEG-Electrodes Positions Measurement System

Abstract

Electroencephalogram (EEG) is a physiological test that measures electric potentials on various locations over the cortex. EEG is wildly used for diagnostics of epilepsy and other brains related disorders. It is also the main method in investigating brain activity in the field of brain-computer interface (BCI).  The electrical activity recorded by electrodes is highly sensitive to the specific location of the electrodes. It is vital in order to perform a precise diagnostics and even more crucial when attempting to solve the inverse problem, in determining which part of the brain controls our intentions in the research of BCI. 
This project presents an automatic method for detection, labeling and the 3D location reconstruction of EEG electrodes in space. Our system will be based on methods of computer vision. It will be low-cost and simple to construct and provide solid results using several images captured by a monocular camera.

Solution Outline

In this project we have designed a system that will produce a 3D reconstruction of EEG electrodes position from several images captured by a monocular camera.
The block diagram of that system is presented below:

 

Electrodes detection:

Automatic detection using color filtering:

 L*a*b color space was chosen  (to minimize the influence of illumination)

 Distribute the n colors with maximum a-b distance

Define robust distance function:                 

Mark electrodes with chosen colors

Empiric threshold value was chosen

Reconstuction & optimization:

Minimize reprojection error of LS problem

 Initialize structure (inital camera matrix & correspondenses ) based on pinhole camera model

An immplemnetation of SBA using off the shelf  “Bundler Toolbox“ (by Noah Snavely - https://phototour.cs.washington.edu/bundler/ )

Error Estimation:

Create an ideal-square-grid, compare vs. the reconstructed one

Estimate the distance between corners by averaging the distances of adjacent points

Reconstruction error:

Experiments and Results

  • Conclusions

    • Downloads