Improved depth estimation algorithm for preserving depth edge and temporal consistency

Depth information acquisition is one of the key technologies of FTV (Free View Television) system. To obtain accurate depth information, an improved depth estimation method is proposed. For each pixel in the reference view and current view, gradients between current pixel and its eight neighboring pixels as well as the average value of the nine pixels´ luminance are computed. To increase depth accuracy especially in areas around depth edges, these gradients and average luminance are combined to compute average luminance-gradient matching errors (ALGME). Both motion information of current frame and depth information of the previous frame are used as a constraint to preserve temporal consistency of the depth map. Experimental results have shown that the proposed method can preserve depth edge and temporal consistency of depth map effectively.