Eksperymentalne stanowisko do integracji i prezentacji widoków przestrzennych

The goal of the project is to develop computer science techniques that enable: the embedding of virtual objects into an image captured by a pair of stereo cameras, taking into account image depth and object occlusion; tracking the viewer’s changing position relative to the 3D scene presented to them; and generating stereoscopic views of the scene tailored to the viewpoint determined by the viewer’s position. An analysis of the stereoscopic image will be conducted to detect depth and automatically select planes within the image. Artificially generated spatial objects will be embedded in the scene prepared in this manner. Systems used for 3D projection, such as a screen and stereo glasses, will be employed. Such a presentation is intended to be personalized for a specific viewer; that is, it will be necessary to track the viewer’s position relative to the virtual object in real time and generate the image accordingly for the viewer’s specific viewpoint.

The project combines the following research topics:

• analysis of stereo image sequences to obtain depth information, followed by segmentation of these images based on the location of objects (separation of planes in the image).
• synthesis of acquired information about the three-dimensional structure of the image and artificially generated meshes in order to place virtual objects in an environment captured by a pair of stereo cameras, taking into account image depth and occlusion of planes.
• real-time tracking of the user using computer vision methods and establishing a relationship between the reference frames associated with the displayed stereo image and the real-world frame associated with the user.
• generating real-time 3D scene views for viewpoints that change over time based on the location of the user (or users) within the scene.
• Processing data from 3D scans for presentation purposes. Such scans are performed with high accuracy, and their processing places significant demands on computer resources. For the purposes of real-time visualization, it may therefore be necessary to dynamically adjust the precision of the object’s mesh representation depending on the viewer’s distance from the virtual object, as well as the direction and level of detail at which the object is to be displayed.

Machine-translated
Przetłumaczone maszynowo