Pupil Detection And3D Depth Detection For Human-Computer Interaction

Autostereoscopic display, which can provide3D images for the observers, is a stereoscopic mode without any accessorial instruments. However, right now people in front of the autostereoscopic display can only watch the3D image at some place. To make the3D images available for observers everywhere, the three-dimensional information of the observers should be detected and traced. In non-intrusive stereo display, the image must be precisely located into the user’s eyes and the distance from the display. This paper makes a main research in the Video Graphic Array, to meet this demand.The main component of our detection algorithm contains:First, taking advantage of Bright Pupil effect with active infrared illumination, we formulate the Candidate screening area, With candidate screening algorithm, we can screen out useful face candidates in a much fast manner. Second, Real Adaboost algorithm for face detection, Real Adaboost algorithm is a high performance robust algorithm for face detection problem. Accompanied by cascaded strategy and tracking algorithm, It can do its job much faster, Haar Based SVM for precise eye locating, our algorithm use Haar features to train the SVM classifier to get a much faster, robust algorithm.In face tracking aspect, a novel algorithm using integral image is proposed. We use template matching algorithm to acquire the precise motion vector between the frames of the sequence. With the aid of the integral image, the time cost is rapidly reduced. Also, we introduce the Novel three step search(NTSS) algorithm, the best matching point can be reached with an acceptable time. Experimental results demonstrate the timing and robustness of the algorithm.Some related concepts of the binocular camera system measuring the depth information are introduced. Both the difficulty of the detection and the autostereoscopic display applications are also described in this thesis in detail. Besides, the main technology of the3d distance detection are introduced and a3d imaging model is given, in the end the experiment process and the results are given.