模式识别在多视角下数据拼接中的应用

针对传统三点定位法在数据配准中使用的局限性，提出了将模式识别中的相似性理论应用在多视角下数据拼接中的新方法．首先，在数据重叠区域内用梯度法搜索曲率极值点，然后在已经获得的曲率极值点中，根据模式识别中的灰度相似性原则确定特征点，最后根据对应的特征点进行配准．将该算法应用在水轮机叶片毛坯的测量数据配准中，结果表明，它可实现空间曲面的多视角测量及后序多片数据的拼接，具有速度快、精度高、可操作性强等特点，而且算法不受零件材质和形状的限制，通用性较强．

Application Study on the Pattern Recognition in the Registration of Measured Data under Different Viewpoints

Aimed at the limitation of the traditional three points locating method in measuring data registration, a new idea applied the similarity theory to pattern recognition to the registration of measuring data under different viewpoints is presented, where the curvature extremum points in data superposition area are searched with the gradient method, then the feature points in the curvature extremum points according to the principle of gay similitude in pattern recognition are acquired, and the measuring data registration is carried out at the corresponding feature points. This algorithm is used in the registration of hydroelectric turbine blades, and the results show that this novel idea can realize the different viewpoints measuring and data registering for 3D surfaces and demonstrates the features of satisfied rate, higher accuracy and flexible operation.