中文摘要：

利用BP神经网络对立体视觉系统进行隐式标定，采用投射大小光斑的方法增加自由曲面的图像特征；提出了一种基于射影变换原理的分块匹配算法，分别在左右两幅图像上，对大点构成的区域进行标准形状变换，然后实现区域内小点的自动匹配．对石膏像面部进行三维重构的实验表明，投射光斑的方法增加了自由曲面的图像特征，分块匹配算法能有效地实现图像间的特征匹配．

英文摘要：

The implicit calibration was applied to the stereo vision system using BP neural network. The image features of a free-form surface were enhanced by means of projecting big and small dots. On the basis of projective transformation, an area matching algorithm was presented. After the shapes of areas formed by big dots were transformed into standard shapes （for example, square） on the left and right images respectively, all the small dots within the standard area were matched automatically with their corresponding partners within the templet standard area. The experiments of 3D reconstruction of plaster model indicated that the dot projection enhances the image features of the free-form surface and the area matching approach performs matching effectively between the images.