Automatic viewpoint selection for polygonal objects through projected mesh analysis

Abstract

The camera position both reveals and hides parts of a 3D object. Additionally, the shape and discernible information of the object vary significantly with the camera’s point of view. While existing research has focused on identifying the best point of view for solid objects, the additional information provided by semi-transparent/translucent objects remains underexplored. This paper introduces a new approach that, without prior knowledge of a polygonal 3D object (whether solid or translucent), can automatically determine its best viewpoint by analyzing what the user sees on the screen: the object projected in 2D. Although identifying the best view may initially seem subjective, this paper demonstrates how this decision can be effectively systematized by a computer. Our method leverages an unsupervised learning approach to discover optimal viewpoints without requiring labeled datasets, thereby automating the process and reducing the need for human intervention. The optimal viewpoints identified by our approach were experimentally compared with those selected by users. Our results indicate that, on average, our method selects viewpoints of higher perceived quality than the users’ manual selections.