Global object placement relation for improving 3-D scene construction from single image

Object Placement Relation (OPR) is the set of properties with classified relations between two connected objects. OPR has been used to predict depth level and locate position of object in 3-D scene. However, since the OPR does not identify the relation of each object to overall scene, it lacks consistency when applied to complex 3-D scenes. The results show distortion and unsmoothness. This research proposes the use of Global Object Placement Relation to improve quality and continuity of the scene. Global OPR is an extended OPR using graph theory to identify relation among objects in the scene. Inner and outer conflicts in OPR classification are identified and eliminated. More steps to increase accuracy in depth estimation are added by measuring each object with the horizon line position, correcting position of occluded objects, and rechecking the result of Octree structure with camera parameters. Our experiment results show increase in accuracy of classification and deliver a more realistic, better quality of 3-D scene.