Surface Feature Extraction from Range Image Based on Grid Sampling

Abstract

The high-level feature extraction from range image is one of the most important areas of 3D computer vision. In order to enhance the processing time and to simplify the structure of the distributed points, a new fast method is presented to transform the irregular distributed 3D point clouds to range data: have z axis represent the depth information, and realize the regular grid sampling along xy axes., This method can be applied to any type of structure light system. Thereafter, a hybrid range image segmentation algorithm combined with edge detection and region growing is described. The whole image edge-map, including the jump and crease edges of the whole image is automatically grabbed. The normal and angle of local surface can be estimated through minimizing the sum of squared Euclidean distance by PCA technique. We have carried out extensive tests using real range image acquired by two range image finders, and the algorithm turns out to be superior to many traditional edge and region methods with regard to the anti-noise and real-time parallel measurement satisfaction. The good results demonstrate that the method fits well in the arbitrary shape of objects and can be used in many computer vision tasks such as 3Dmodel reconstruction, robotic independence navigation, reverse engineering, digitizing historic sites, etc.