Distinguishing objects on point cloud data

For many years, a variety of studies have been sought in the problems of distinction of objects, in computer graphics and computer vision areas. Although there are many image-based approaches to solve this problem, it is thought that these solutions are insufficient to be used in some areas. Distinguishing and recognizing between objects and shapes that are frequently encountered in everyday life is an important question in these areas. In this thesis, a curve skeleton extraction algorithm is applied to the point clouds of objects used in real life in order to distinguish the physical objects in a scene. The curve skeleton of the 3D object is an abstract, geometric and topological representation of 3D shapes. This skeleton structure is the integrated state of an average point cloud data obtained from the existing point cloud. Medial skeleton extraction algorithm and its assistant Laplace shrinking algorithm are applied for extracting curve skeleton information on point cloud data. There is a data type and a method of operation that each of the existing skeletal extraction algorithms uses in its own way. The most basic way of achieving a robust skeletal structure is that the point cloud of objects is being as perfect as possible. Its way is to implement a good scanning of the objects. There are a variety of applications in this area, such as the use of multiple or moving sensors in the scanning process. In this thesis, an approach which generates a scan with a natural movement of the user for the object is implemented by means of a fixed Kinect sensor, which is more practical and cost-effective than conventional scanners. This approach allows the production of indoor and short-range 3D objects in a wide range of full (360 degrees) point cloud data. The system for object scanning is easy to set up, producing simple and impressive results. In response to a single fixed Kinect sensor, the 3D object standing on the turntable is rotated at specific angles (eg. 90) to obtain multiple point cloud scan data. All the point clouds are shifted so that the center of gravity of each scan is (0, 0, 0) for merging and aligning operations. The 3D rotation process is performed in the y-axis direction with respect to the center point (0, 0, 0) for the point cloud data obtained from the second and subsequent scans. The transformed point cloud data obtained from the different angles are aligned with respect to each other, shifted relative to the specified junction key points. An algorithm running on the cross sections for the improvement and reduction processes is carried out on the obtained full 3D coarse point cloud data. Thus, the obtained scan is free from the data and has a 3D, simple and regular structure. In this respect, it has been ensured that the point clouds are optimized to allow skeleton extraction.