PROJECTS AT VISL FINISHED IN 2008


Eyelid marking tool

by Danny Shchupak
Supervised by Johanan Erez


Abstract

In this project, an application was built for marking eyelids in eye images from an IR video camera. The application shall be used in order to manually build a data base of marked eye lids. This data base shall be used as "ground truth" data for comparing and testing automatic real-time eye lid marking algorithms.
The project is part of a research in association with Elbit Systems Ltd.


Demands

The utility should be capable of automatically detection of eyelid contours, relying on the manual marking by the user. It has to be able to deal with minor inaccuracies and not to be affected due to poor marking of the user. The detection shall be accomplished in a reasonable time frame. The effectivity of the tool will be tested on a variaty of images, and the results be saved in a database.


The Solution

It was chosen to use the algorithm of the "Intelligent Scissors" project, previous held in the VISL. The tool that was developed employs automatic detection, based on user's input, using several techniques of edge detection. After applying these techniques on the image, the weight matrix is created and is used as input for Dijkstra algorithm (see documentation of the "Intelligent Scissors" project). Finally, the contour line is added to the picture, and can be saved in the database.

Tools

The tool was developed in MATLAB 2007b, the eye images were taken by ELBIT's eye tracking lab.

Results

The tool showed capability for accurately draw the eyelid contour at some examples, however the success depends on the accuracy of the user input and the quality of the input image. Runtime is about 2-3 minutes for each image.
Improvements have to be made to get more robust results, maybe integrate other methods and improve runtime.

Acknowledgment

I am grateful to the project supervisor Johanan Erez for his help and guidance throughout this work.
I am also grateful to ELBIT Systems Ltd. and the Ollendorf Minerva Center Fund for supporting this project.

Related documentation