Abstract: To avoid camera calibration in the implementation of human-computer interaction system with large screen based on binocular stereo vision, one approximate and easy method is proposed. The distance between two cameras, left camera and the left side of screen, cameras and floor, the bottom side of screen and floor are measured to find the point of floor which is imaged as the same point with fingertip based on perspective projection model. One camera is connected with the point of floor as one space line. Two lines are intersected at the three-dimension spatial coordinate of fingertip served to position two-dimensional coordinate in the large screen. Mouse click is simulated by bend and stretch of index finger achieved by fingertip detection algorithm. Users are not aware of error even if position pointed by finger is inaccurate because mouse is seen by users when they move the hand. The method proposed is easier and more convenient than calibration method due to inaccurate position not seen by users so that system can be rapidly accomplished.

Key words: large screen, human-computer interaction, binocular vision, perspective projection

摘要： 针对利用双目立体视觉技术实现大屏幕人机交互系统时摄像头标定复杂的问题, 提出一种大屏幕人机交互系统的实现方法。由两个摄像头负责采集用户手部的图像，获取指尖的二维图像位置坐标，通过透视投影模型，由测量所得的两个摄像头之间的距离、大屏幕左边与摄像头之间的距离、摄像头离地面的高度、大屏幕底边离地面的高度等数据，求出地面上与指尖成像在同一位置的地面坐标，由摄像头的位置及地面坐标求出一条直线，两条直线相交获取指尖空间坐标，由指尖空间坐标选取与屏幕对应的二维平面坐标，经物理坐标到逻辑坐标转换求出手指指向屏幕的实际坐标，实现大屏幕的定位。定位完成后，采用指尖检测方法检测视频图像中指尖是否存在，以此判断食指弯曲和伸开的点击操作。指尖空间坐标的计算方法虽然存在误差，但由于用户在操作过程中能够实时看到鼠标的位置，因此该误差对用户是透明的，而这种方法以简单的、近似的求解避免了复杂的摄像头标定过程，便于系统工程的实施。

关键词: 大屏幕, 人机交互, 双目视觉, 透视投影