nonoccluded

Nonoccluded is an adjective used in computer vision and related fields to describe objects or parts of a scene that are fully visible and not obstructed by other objects or by the object's own geometry. A nonoccluded object can be observed without occlusion-induced ambiguity in its shape, texture, or pose, which simplifies tasks such as detection, segmentation, tracking, and 3D reconstruction.

In practice, many real-world scenes include occlusions, so nonoccluded instances are often selected for ground-truth annotations or used as idealized data to evaluate algorithms. Datasets may label instances as nonoccluded when the entire object is visible, or provide visibility annotations such as a visible fraction or occlusion mask.

Applications of nonoccluded data include training and evaluating models for object detection and pose estimation under unconstrained visibility, and improving the accuracy of 3D reconstruction where full geometry is observable. In robotics and augmented reality, nonoccluded views can facilitate reliable grasp planning, scene understanding, and rendering when complete visibility is available.

Challenges arise because occlusion is common in natural environments, making nonoccluded data relatively rare. Techniques to address occlusion include multi-view fusion, temporal tracking across frames, depth sensing, and the use of priors or shape completion to infer hidden surfaces when objects are partially obscured.

See also: occlusion, occluded, partial occlusion, visibility, mask.