Consistent depth maps recovery of video via object segmentation
Journal
ACM SIGGRAPH 2013
Date Issued
2013
Author(s)
Abstract
Video depth reconstruction is a long-term problem in computer vision. A fair amount of previous work adopts Markov random field model in depth estimation formulation. Various local prior terms are included into global smoothness of those optimization frameworks. However, the estimated depth discontinuities are usually not consistent with object boundaries and intra-object smoothing is not well achieved. Such problems are addressed in color segment-based methods but cannot be solved well when color segments exist across object boundaries or are not temporarily consistent.
Type
conference paper