基于几何特征和图像特征的点云自适应拼接方法

Adaptive Point Cloud Registration Method Based on Geometric Features and Photometric Features

多视点云拼接技术是物体三维测量过程中的重要环节。现有的无标志点三维点云自动拼接方法在对不同表面进行测量拼接时稳定性较差。针对此问题,提出了一种基于几何特征和图像特征的点云自适应拼接方法。该方法建立了一个配准算法选择模型,通过引入配准算法判断因子来综合评价物体表面的几何、纹理复杂程度,从而系统可根据判断因子自适应地选择合适的配准算法,实现基于几何特征配准和基于图像特征配准的有机结合。并在特征点匹配过程中,采用随机抽样一致(RANSAC)算法对误匹配特征点进行剔除。实验结果表明,该方法可实现不同表面的稳定点云拼接。

Multi-view data registration is an important step in the process of large objects three-dimensional （3D） measurement. But the available unmarked 3D surface auto-registration methods can result in unstable registration results when measuring objects with different surface feathers. Aiming to solve this problem, an adaptive 3D auto- registration algorithm is presented based on both geometric and photometric features. In this algorithm, a registration selection model is built to generate a registration judgment factor for synthetically evaluating the complexity of surface geometry and texture. Based on this model, an appropriate registration strategy can be adaptively selected to promise a reliable registration result. Moreover, random sample consensus （RANSAC） algorithm is used to remove the remaining wrong correspondence. The experiments use various registration results to illustrate the performance of the proposed method in different measurement applications.