Adaptive Selection of Motion Models for Panoramic Video Coding

Abstract

A panoramic video is an image-based rendering (IBR) technique which provides users with a large field of view (e.g. 360 degree) on surrounding dynamic scenes. It includes not only the translational motions but also the nontranslational motions, such as zooming, rotation and uneven stretching etc. This paper presents a motion compensated prediction scheme based on adaptive selection of motion models to predict the complex changes between successive frames efficiently in panoramic video coding. By performing the initial motion estimation phrase and the refined motion estimation phrase in the proposed scheme, simulated results show that the coding performance of the proposed scheme is much higher than the traditional motion compensated prediction scheme in panoramic video coding.

A Survey on 360-Degree Video: Coding, Quality of Experience and Streaming

“…The spherical coordinate transform can be used to further improve performance and extend the possible motions to the whole 3D space [82], working in spherical coordinates and using relative depth to convert between ERP and the 3D space. It is also possible to assign different motion vectors to pixels in the same block, correcting the motion vector distortion [83]. A less efficient but less computationally demanding way to correct motion vectors in ERP is to exploit the WS-PSNR [84] weight map to calculate a scaling factor for the motion vectors [85].…”

Deformable block-based motion estimation in omnidirectional image sequences

“…Partitioning the frame in blocks of variable sizes [4] or using high-order motion models [5], [6] have been proposed as solutions to overcome the limitations of this model, at the price of an overhead to signal the partitions and increased complexity. Such solutions may also improve motion estimation in omnidirectional videos, as demonstrated in [7], where a rate-distortion optimal selection of translational, affine or quadratic motion model has been investigated for panoramic videos resulting from cylindrical or equirectangular map projection. Nevertheless, they do not take into account the spherical camera geometry and the map projection, thus, for example, they are unable to handle discontinuities.…”