Acquisition, representation and rendering of real-world models using polynomial texture maps in 3D

Abstract

The ability to represent real-world objects digitally in a realistic manner is an indispensable tool for many applications. This work proposes and evaluates a method for acquiring, processing, representing, and rendering these digital representations. Besides the 3D geometry of the object, which defines its shape, and the texture, which defines how the surface looks, this work highlights the importance of capturing also the reflectance behavior information of the object. This defines how the surface of the object looks when being lit from different directions. The proposed method makes it possible to acquire all the required data using a straightforward, non-expensive method, using only Microsoft’s Kinect sensor, three reflective spheres, and a light source. The 3D geometry is acquired using the Kinect’s depth sensor, by capturing data from different viewpoints. The data is then processed through a pipeline in order to create a single, registered point cloud. Texture and reflectance behavior information are acquired by lighting the object from different directions and capturing data using Kinect’s RGB camera. This information is then processed and represented using multiple overlapping Polynomial Texture Maps. Finally, this work extends the use of Polynomial Texture Maps to 3D, by examining how the registered point cloud and the overlapping Polynomial Texture Maps can be mapped together to produce the digital representation. This allows examination of the model, as well as interactively changing the viewpoint and also light direction in real-time. Varying light direction uncovers details of the object which would not have been possible to observe using a single, fixed, light direction. The obtained results are very positive, and show the proposed method of representing real-world objects, which uses geometry, texture and reflectance behavior information, produces realistic representations and discloses more object features than other available representation methods. These results render this work useful in many scenarios, amongst which is the examination of cultural heritage artifacts with surface variations.