一种适用于PC机的视觉注意点定位算法

人眼观察屏幕时,其脸部姿态与屏幕视觉注意点具有内在对应关系。摒弃传感器、头盔等跟踪设备,提出并研究一种适用于日常PC机的视觉注意点位置获取方法。使用摄像头抓取人脸实时图像,经Haar级联分类器提取能够反映脸部姿态的特征:双眼、口、鼻等,通过BP神经网络训练出特征相对坐标与视觉注意点之间的对应关系。实验证明该方法在无需任何专门设备支持的条件下,即能够检测视觉注意点位置,检测精度达到0.1视距。为立体影像或虚拟场景的裸眼立体观察,提供了一种新的解决方式及方法。

A Sight Point Localization Algorithm for Personal Computers

When observing a screen, there is an internal corresponding relationship between the face posture and sight point on the screen. This paper proposes and studies a sight point obtaining method used in personal computers without tracking devices. The camera captures real-time images of the face, and features(the eyes, mouth and nose) of the face are extracted by the cascade classifier. The corresponding relationship between the position of the sight point and the relative coordinates of the feature is trained by BP neural network. A large number of experiments prove that this method is practical, the relationship of facial features and sight point position is reliable. Detection accuracy reaches 0.1. This method provides a new method for the naked eye observation of stereoscopic image or virtual scene.