周喆, 沈建新, 韩鹏, 江俊佳. 基于Census变换和引导滤波的立体匹配算法[J]. 应用光学, 2020, 41(1): 79-85. DOI: 10.5768/JAO202041.0102003
引用本文: 周喆, 沈建新, 韩鹏, 江俊佳. 基于Census变换和引导滤波的立体匹配算法[J]. 应用光学, 2020, 41(1): 79-85. DOI: 10.5768/JAO202041.0102003
ZHOU Zhe, SHEN Jianxin, HAN Peng, JIANG Junjia. Stereo matching algorithm based on Census transformation and guided filter[J]. Journal of Applied Optics, 2020, 41(1): 79-85. DOI: 10.5768/JAO202041.0102003
Citation: ZHOU Zhe, SHEN Jianxin, HAN Peng, JIANG Junjia. Stereo matching algorithm based on Census transformation and guided filter[J]. Journal of Applied Optics, 2020, 41(1): 79-85. DOI: 10.5768/JAO202041.0102003

基于Census变换和引导滤波的立体匹配算法

Stereo matching algorithm based on Census transformation and guided filter

  • 摘要: 在双目立体视觉系统中,立体匹配是关键步骤之一,其精度对后续的研究有着重大影响。Census算法由于具有简单明晰、运行效果好、实时性强等优点,被广泛采用。但Census立体匹配算法存在变换窗口中心点易受外界条件干扰、深度不连续区域匹配精度低等缺点,由此提出了一种新型的基于Census变换及引导滤波器的立体匹配算法。在Census变换阶段通过计算变换窗口周围的像素的平均值,降低了外界干扰的影响,同时在代价聚合阶段引入具有包边特性且计算量不依赖于滤波核大小的引导滤波器作为自适应权重。实验结果表明:所提算法在Middlebury测试平台上平均误匹配误差为6.03%,相较于目前Census立体匹配算法16.2%的平均误匹配率,匹配效果明显提高,且算法效率较高,具有较好的辐射不变性。

     

    Abstract: Stereo matching is one of the key steps in binocular stereo vision system, and its accuracy has a significant impact on the subsequent research. The Census algorithm is widely used because of its advantages of simplicity and clarity, good running effect and strong real-time performance. However, the Census stereo matching algorithm also has some disadvantages, such as the center point of transform window is easily disturbed by the external conditions and the accuracy of the depth discontinuous region is low. Therefore, a new stereo matching algorithm based on Census transformation and guided filter was proposed. In the stage of Census transformation, the impact of the external interference was reduced by calculating the pixel average around the transform window. And at the same time, in the stage of cost aggregation, the guided filter with tipping characteristics and its computational complexity was independent of the size for filter nuclear was introduced as the adaptive weight. The experimental results show that the average mistake matching error of the proposed algorithm is 6.03% on the Middlebury test platform. Compared with the current average mistake matching rate of 16.2% for Census stereo matching algorithm, the matching effect is obviously improved, and the algorithm has high efficiency and good radiation invariance.

     

/

返回文章
返回