Omnidirectional Images Overview

• Omnidirectional images are images with a 360° field of view captured using equirectangular projections, enabling immersive visualization.
• They pose challenges such as severe projection distortions, non-Euclidean geometry, and sampling inhomogeneity that require advanced processing techniques.
• Applications in VR, robotics, and social network services demand seamless visual continuity and precise computational methodologies for effective deployment.

An omnidirectional image (ODI) is an image whose field of view covers the entire sphere around the camera, typically encoded using an equirectangular projection (ERP) at a 2:1 aspect ratio. ODIs serve as canonical data in immersive applications that require perception or interaction with the full visual environment, such as virtual reality (VR), robotics, and social network services. The acquisition, representation, processing, and generation of ODIs introduce unique geometric, algorithmic, and psychoperceptual challenges, owing to the non-Euclidean nature of the sphere, severe projection-induced distortions, sampling inhomogeneity, and the need for seamless visual continuity. This article surveys the technical underpinnings, core computational methodologies, quantitative benchmarks, and contemporary directions in the research and application of omnidirectional images.

1. Mathematical Representation and Projections

An ODI models the radiance field surrounding a point in space, such that any viewing direction can be mapped to color (and, optionally, depth or semantics). Formally, the