Affordable Access

Authors
Publisher
Institut for elektroniske systemer
Publication Date
Disciplines
  • Mathematics

Abstract

In the past few years personalised advertisements have become a popular method of attracting attention, especially on the Internet. Transferring this concept to real world shops gave the idea of an interactive system. This should tell what a potential customer in front of the shop window is looking at, and then choose an appropriate commercial to show to this customer.<br/><br/>During this project a method for estimating the head pose of a person has been proposed and tested. This system makes use of OpenCV's frontal face detector, in order to find the region of interest in the images. For the head pose estimation a geometric approach is tried out, where three-dimensional coordinates for three facial features are found and used in the calculation of the head pose. In order to obtain three-dimensional coordinates two mid-range web cameras have been employed in a stereo set-up, from which the depth of the facial feature points can be calculated from disparity. To locate the facial features, eyes and mouth, three different methods have been tested, all based on Haar-like features. The final method makes use of an attentional cascade boosted by AdaBoost.<br/><br/>Testing the system showed that all three facial features are detected in 90 \% of the images, but the locations of the facial features are not exactly the same in the two frames from the stereo cameras, introducing a problem for the disparity calculations. The stereo vision and pose calculations are tested by manually choosing the facial feature points. This showed that with the current resolution of the cameras the pose calculations can not be done with enough precision for the distances specified for the application. Tests at shorter distances showed that the desired precision can be obtained with a higher image resolution.<br/><br/>The outcome of the tests indicates potential improvement and with further development of the feature detection and better equipment the approach seems applicable.<br/>

There are no comments yet on this publication. Be the first to share your thoughts.

Statistics

Seen <100 times
0 Comments