Kinect三維資料的超高解析度重建

Super-resolution Reconstruction for Kinect 3D Data

邱于萍(Yu-Ping Chiu)

指導教授 ： 柳金章

國立中正大學/工學院/資訊工程學系/碩士(2013年)

摘要

在一些應用環境下，我們只能取得低解析度的Kinect三維資料。為了要獲得高解析度Kinect三維資料，我們需要進行Kinect三維資料的超高解析度重建，而超高解析度重建是一個ill-posed 問題。在本研究中，針對Kinect三維資料，我們將提出對應的超高解析度重建方法，其中Kinect三維資料包含一張高解析度彩色影像及一張低解析度深度影像。由於從Kinect拍攝的低解析度深度影像在物體邊界有遮蔽問題，利用輔助的高解析度彩色影像，本研究首先填補深度破洞，之後將深度影像放大以得到高解析度深度影像。放大後的高解析度深度影像可能在物體邊界會有不正常的現象導致深度影像不正確。本研究提出在物體邊界做區域性改善，利用高解析度彩色影像的分割圖來改正邊界以得到更準確地高解析度深度影像。根據實驗結果顯示，本研究所得到的高解析度深度影像比其他四個比較方法得到的高解析度深度影像更準確。

關鍵字

深度影像 ； 超高解析度Kinect 三維資料重建 ； Kinect 三維資料

並列摘要

In some application environments, only low-resolution (LR) Kinect 3D data are available. To obtain high-resolution (HR) Kinect 3D data, it is required to perform Kinect 3D data super-resolution reconstructions, whereas super-resolution reconstruction is an ill-posed problem. In this study, a super-resolution approach for Kinect 3D data is proposed, where Kinect 3D data include an HR color image and an LR depth map. Because the LR depth map from Kinect has the problem of occlusion near object boundary, the proposed approach fills the depth hole first and then upsamples it to get the high resolution depth map using the assisted HR color image. Edge of objects may be some artifacts of upsampled HR depth map and the depth map is not accurately. The proposed local edge enhancement can rectify the edge using the segmentation map from HR color image to obtain more accurately HR depth map. Based on the experimental results obtained in this study, the final HR depth maps of the proposed approach are better than those of the four comparison approaches.

並列關鍵字

depth map ； Kinect 3D data super-resolution reconstruction ； Kinect 3D data

參考文獻

[4] W. C. Chiu, U. Blanke, and M. Fritz, “Improving the Kinect by Cross-Modal Stereo,” in Proc. of the British Machine Vision Conf., 2011, pp. 116.1-116.10.

[1] F. Qi et al., “Structure guided fusion for depth map inpainting,” Pattern Recognition Letters, vol. 34, no. 1, pp. 70-76, Jan. 2013.

[2] D. Miao. et al., “Texture-assisted Kinect depth inpainting,” in Proc. of IEEE Int. Symp. on Circuits and Systems, 2012, pp. 604-607.

[5] S. Milani and G. Calvagno, “Joint denoising and interpolation of depth maps for MS Kinect sensors,” in Proc. of IEEE Int. Conf. on Acoustics, Speech, and Signal Process., 2012, pp. 797-800.

[7] N. E. Yang, Y. G. Kim, and R. H. Park, “Depth hole filling using the depth distribution of neighboring regions of depth holes in the Kinect sensor,” in Proc. of IEEE Int. Conf. on Signal Process., Communication, and Computing, and Computing, 2012, pp. 658-661.