人体表面特征点三维空间坐标的实时监测是实现自然高效、非接触的智能化人机交互的关键技术。算法在提取肤色区域外轮廓的基础上, 依据手指和手势的多种形状约束特征实现指尖定位。基于Adaboost算法确定人脸区域, 针对是否佩戴强干扰性眼镜, 采用不同的灰度及空间结构特征实现人眼中心的准确定位。对左右内眼角点或是黑框眼镜内框曲率点进行改良性的光流跟踪。通过多尺度逐级定位算法提升特征点的定位精度。基于双目立体视觉原理实现特征点的三维空间坐标重构, 并依靠三维信息反馈有效重启跟踪错误的光流算法。算法将特征与统计方法相结合, 通过多尺度以及多信息(灰度、色调、空间结构、三维)的综合运用, 相对于传统算法进一步提升了算法的时间处理效率和复杂背景中多状态下(强干扰佩戴物、较大角度的姿态变化、遮挡、光照)目标特征点的检测正确率。

Three dimensional space coordinate real-time measurement is one of key and basic techniques in non-contact、high efficient and natural man-machine interaction. Algorithm can locate the tip of finger based on shape restrain feature of finger and hand gesture on the contour of skin hue region. Face region is recognized by Adaboost algorithm. According to whether wearing strong interference glass, different algorithm based on gray scale and space structure are used to accurately locate the center of eyes. The light trace algorithm is improved by tracing the certain feature points (inside corner points of left and right eyes or inside frame curvature points of black frame glass ). Algorithm increases the location accuracy of feature points by multiscale step location method. The three dimensional coordinate of feature points is reconstructed based on binocular stereo vision, and the light trace with error can be restarted by three dimensional information feedback. Algorithm combines feature and statistics methods, synthetically applies multiscale and multi-source information. Compared to traditional algorithm, the time efficiency and the accuracy ratio of object feature points recognization in multiple situation ( strong interference wearing, gesture change in big angle, shelter, illumination) in complex backgroud picture are increased.