Research Of Human-Computer Interaction Algorithm Used In Stereoscopic Display

This paper describes important machine-learning algorithms implementing an interactive three-dimensional display system and the main algorithm mainly includes three parts:face detection, pupil location and gesture recognition. The first part detects the position and size of users’face in successive video frames, providing initial candidate areas to locate the pupils. The second part services for stereoscopic display and locates the pupils in video frame images. Based on this, the directivity grating projectes parallax image precisely onto the pupils and ensures users to experience stereoscopic display without 3D glasses. The third part provides users the conctolling of the images displayed on the screen. It detects the position of the user’s palm, records the track of users’ gesture, and then, responds to users after recognition.Face detection algorithm selects candidate face region using skin detection method. This paper proposes a "feature binding" approach for face occlusion such as wearing a hat or a mask. It picks up the "hot zone" of the standard faces and improves the structure of the waterfall cascade by adjusting the arrangement of weak classifiers. Therefore, the detection rate is improved while the speed detection is accelerated. Haar features extract the gradient information of the standard human face.11 waterfall cascade structure accelerates the exclusion of non-face and reduces the false detection rates.Pupil localization algorithm uses the results of face detection and the geometric relation of standard face to extract pupil candidate regions.3 waterfall cascade structure is used for excluding non- eyes in the human eye detection and SVM algorithm for accurate detection. Besides, this paper proposes the use of Kalman smoothing filter for the mutation of the pupil location. Therefore, the location of pupil is more stable.Gesture recognition algorithm locates the user’s right hand with skeleton model of Kinect. After a simple smoothing and denoising, we design an improved 8-derection coding toquantify the gesture trajectory feature into sequence of observations. Considering the high demand of gestures for 3D interactive systems, this paper designs a codec-manner gesture recognition to segment complex gesture. It is also reduces storage consumption of system model and ensures real-time gesture recognition. Correspondingly, a two dimensional filter and inflection point detection is proposed to segmentation the gestures to be recognized. For each sub-gesture, Hidden Markov model is used identify the classification. Moreover, in order to show the result of entire interactive three-dimensional display system, a 3D cube is designed to be controlled by users’ gestures.