In order to transmit real-time video, a CCD camera was implemented as the transmission medium. To process the input facial image, an MPEG file was created using DirectShow SDK, which supports multimedia streaming in MicroSoft Windows. Based on the 20 input feature points, the location of the feature point in each frame was based on the location of the same feature point in the previous image.

The two-dimensional facial features were tracked using a template matching method using a coarse-to-fine tracking method that changes the template size in the search area of this particular region. The coarse-to-fine tracking method is one of the registration strategies that performs registration while increasing the resolution. When performed in real time, a transition image with appropriate resolution can be obtained within the desired time. However, if the data is acquired incorrectly at low resolution, there is the potential for errors to propagate. Matching refers to the task of selecting candidate feature points in the second image that can correspond to a particular feature point in the first image.

Using the 2D coordinates tracked from the right and left images, 3D coordinates could be obtained using stereo image tracking points. The 3D data obtained allowed the creation of a compatible file in 3D Studio MAX, a 3D package.

Keywords

artificial intelligence, facial features, Face recognition, DirectShow structure and feature points