Automatic generation of point cloud synthetic dataset for historical building representation

3D point clouds represent a structured collection of elementary geometrical primitives. They can characterize size, shape, orientation and position of objects in space. In the field of building modelling and Cultural Heritage documentation and preservation, the classification and segmentation of point clouds result challenging because of the complexity and variety of point clouds due to irregular sampling, varying density, different types of objects. After moving into the era of multimedia big data, machine-learning approaches evolved into deep learning approaches, which are a more powerful and efficient way of dealing with the complexity of semantic object classification. Despite the great benefits that such approaches brought in automation, a great obstacle is to generate enough training data, which are nowadays manually labeled. This task results time-consuming for two reasons: the variety of point density and geometry, which are typical for the Cultural Heritage domain. In order to accelerate the development of powerful algorithms for CH point cloud classification, in this paper, it is presented a novel framework for automatic generation of synthetic dataset of point clouds. This task is performed using Blender, an open source software which permits to access to each point in an object creating one in a new mesh. The algorithms described allow to create a great number of point cloud synthetically, simulating a virtual laser scanner at a variable distance. Furthermore, these two algorithms not only work with a single object, but it is possible to create simultaneously many point clouds from a scene in Blender also with the use of an existing model of ancient architectures.