Depth information can be obtained from 2-dimensional (2D) resources through 2D -to-3D conversion technology,in order to meet the demand for 3D content.In this paper,a novel approach based on nonlocal random walk (N RW) and motion compensation is proposed to solve the problem of depth map optimization in depth estimation.The depth maps of key frames and non-key frames are obtained on the basis of NRW and shif ted bilateral filtering (SBF),respectively.In order to sharpen the depth boundaries,taking texture infor mation into account,NRW is utilized to optimize the depth sequence. Considering that the temporal information of adjacent frames,motion compensation is introduced to further improve the quality of dept h maps of non-key frames.Experimental results show that,in contrast to the method directly proce ssed by SBF,our method can produce higher quality depth sequence with accurate object boundaries.