Infrared Eye Detection

Software

The contrast of the camera is increased so that the pupils will stand out in the resulting image. This will also cause other white areas to "flatline" at intensity 255, but will greatly assist in thresholding when trying to "pull" the eyes out of the image later. For people with lighter color eyes, the area around the pupil is still dark because of the contrast adjustment. This is absolutely one of the areas that required quite a bit of tweaking to get right. There are still adjustments to make for people with extremely light colored eyes, or maybe even contact lenses.

Detection of eyes within an image is done using a threshold to remove objects not within the desired intensity range. The image is then segmented and an object list is generated containing the coordinates and size for each potential eye. Every object pair is compared based on size and location. The criteria for matching objects to be eye pairs in this implementation are similar sizes, within a certain distance apart, and the angle of the eyes is mostly horizontal. Object pairs that pass all of these criteria are then considered to be a pair of eyes. The point half-way between the eyes is then estimated to be center-of-mass for a face.