基於十字聚合區域的雙眼立體匹配演算法

Optimization of Stereo Matching Based on Cross-Based Local Construction

陳婉君(Wan-Jun Chen)

指導教授 ： 貝蘇章

國立臺灣大學/電機資訊學院/電信工程學研究所/碩士(2016年)

https://doi.org/10.6342/NTU201603235

摘要

近年來，立體視覺應用於許多領域，如數位媒體,數位醫學，物體檢測，航空，
建模等等方面物體檢測以及建模等等。通常，深度圖和其對應的彩色圖像用於重
建3D場景。如何獲得即時的高精度的深度圖是立體影像研究中極為關鍵的一點。
本篇論文提出的演算法是基於Semi-Global Matching演算法的，並以動態生成的
十字區域作為聚合區域。該提出的演算法在準確性和速度上達到了極好的平衡
（中和）並需要較少的硬體消耗。

關鍵字

立體匹配 ； 十字聚合 ； 半全域匹配

並列摘要

Recent years, stereo visions are widely applied to various elds，such as digital
entertainment, digital medical, obstacles detection, aerospace and modeling.
Usually,depth maps and corresponding color images are used to reconstruct
3D scenes. How to achieve high-accuracy depth maps in real time have be-
come the key point in stereo vision study.
The algorithm proposed in this article is based on the Semi-Global Matching
algorithm using adaptively upright cross support region as cost aggregation
area. The proposed method have gained rather good balance between acc-
uracy and speed also with reduced hardware consumption.

並列關鍵字

stereo matching ； cross-based aggregation ； semi-global matching

參考文獻

ateral filtering. In Data Processing, Visualization and Transmission, pages 455-

[3] T. Federico and et al. Classification and evaluation of cost aggregation meth-

ods for stereo correspondence. In Computer Vision and Pattern Recognition, p-

[4] H. Heiko. Stereo processing by semiglobal matching and mutual information.

Transactions on pattern analysis and machine intelligence, 30(2):328-341, 2008.