利用十字視窗與區域視差精煉之立體匹配演算法

Improved Stereo Matching with Cross-Based Windows and Region-Based Disparity Refinement

王博生(Bossun Wang)

指導教授 ： 楊士萱

國立臺北科技大學/電資學院/資訊工程系研究所/碩士(2014年)

https://doi.org/10.6841/NTUT.2014.00813

摘要

可適性支持權重(adaptive support-weight, ASW)是區域性立體匹配最具代表性的演算法。由於ASW利用固定視窗內的顏色與空間的相似度做為權重，若各匹配視窗內容重複樣式多則容易造成深度估計錯誤。本論文提出視窗內輪廓分布判斷方式，在輪廓分佈較為稀疏之區域使用視窗內各點顏色相似性，形成一個判斷空間相似度的十字視窗配合條件式權重估計深度；在輪廓分佈較為密集之區域使用ASW演算法。為了提升深度估計的準確性，本論文更進一步使用視差精煉(disparity refinement)的方式提升準確率，對左右張深度圖進行檢查(left-right consistency check, LRC)，再針對LRC偵測不一致之錯誤點進行區域連通，統計區域內各深度值的比例，依照深度比例填補錯誤區塊。實驗結果顯示本論文可大幅改善ASW空間相似度與固定視窗的問題，平均錯誤率由ASW演算法的錯誤率6.67%，經過本論文改進演算法後降低到5.57%。

關鍵字

Adaptive Support-Weight ； 立體匹配 ； 深度估計

並列摘要

Adaptive Support-Weight(ASW) is the most representative algorithm of local stereo matching. ASW calculates support-weights of the pixels in a given support window based on color similarity and geometric proximity. When repeated similar patterns occur in matching windows, wrong disparity estimation may be made. This paper proposed using contour judgment to solve the problem in fixed windows. We build cross-based windows combining a weight condition to replace fixed windows and geometric proximity on ASW in the contour area, while the ASW algorithm is still applied in the contour area. To increase accuracy, this paper uses the Left-Right Consistency Check (LRC) method to check the inconsistent areas in disparity map. We fill the inconsistent areas according to the distribution of disparity values. Our results show that the proposed method improves the performance of ASW. The average error rate reduced from 6.67% to 5.57%.

並列關鍵字

Adaptive Support-Weight (ASW) ； Stereo Matching ； Depth Estimation

參考文獻

[2] K. Yoon and S. Kweon, “Adaptive support-weight approach for correspondence search,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol.28, no.4, pp.650-656, 2006.

[4] F. Tombari, S. Mattoccia, and L. Di Stefano, “Segmentation-based adaptive support for accurate stereo correspondence,” PSIVT, 2007.

[5] K. Zhang, J. Lu, and G. Lafruit, “Cross-based local stereo matching using orthogonal integral images,” IEEE Trans. Circuits and Systems for Video Technology, vol.19, no.7, pp.1073-1079, Jul. 2009.

[6] D. Comanicu and P. Meer, “Mean shift: A robust approach toward feature space analysis,” IEEE Trans. Pattern Analysis and Machine Intelligence, vol.24, pp.603-619, May 2002.

[7] P. Meer and B. Georgescu, “Edge detection with embedded confidence,” IEEE Trans. Pattern Analysis and Machine Intelligenc, vol .23, pp.1351-1365, Dec. 2001.