A 3D video quality evaluation model that provides high depth perception satisfaction and efficient transmission channel use based on depth cues

In the field of Computer Vision, where scientific research and academic studies are accelerating, the popularity of applications based on the use of 3 Dimensional (3D) videos is increasing day by day. There are many variables that determine the quality of 3D videos such as indoor-outdoor differences and environmental factors. In this context, it is essential to develop new methods for quality evaluation of 3D videos to be used in modern applications to provide high depth satisfaction and efficient transmission channel use. In this thesis, a dynamic, adaptive, and No Reference (NR), which does not require any reference video, 3D video quality assessment model is proposed that aims to increase depth perception satisfaction in viewers by taking into account the Human Visual System (HVS) characteristics. The proposed model is based on measurement methods derived from the color images and/or the depth maps that make up the 3D videos. In this study, where utmost importance was given to the use of deep learning-based approaches, Pix2Pix model was used for the Aerial Perspective (AP) measurement method, RAFT model was used for the Motion Information (MI) measurement method, and YOLOv8 model was used for the Retinal Image Size (RIS) and Convergence (CONV) measurement methods. Although all methods are performed on color videos, depth maps are also included in the calculations in RIS and CONV methods. As a result of the studies conducted, all outputs of the proposed methods were obtained to fulfil the expected results and the thesis work was successfully completed.