基于改进SGM的森林场景视差图生成方法  

Method for generating disparity map of forest scene based on improved SGM

作　　者：李智 李兴东 LI Zhi;LI Xing-dong(School of Mechanical and Electrical Engineering,Northeast Forestry University,Harbin Heilongjiang 150040,China)

机构地区：[1]东北林业大学机电工程学院,黑龙江哈尔滨150040

出　　处：《林业机械与木工设备》2024年第4期31-36,共6页Forestry Machinery & Woodworking Equipment

基　　金：国家重点研发计划项目(2022YFC3003002)。

摘　　要：生成精确的视差图是采用双目视觉对森林冠层下进行三维重建、资源调查的基础。由于森林环境内结构复杂,视差不连续处较多,现有的算法应用于森林场景中会出现误匹配率较大、树干边缘轮廓不清晰的现象。针对上述问题,提出一种改进的Semi-Global Matching(SGM)算法。首先在立体匹配的初始代价计算阶段,为避免由中心像素值突变造成的匹配错误,将Census中心像素值替换为四邻域像素点的平均值。然后对双目相机图像进行k-means聚类分割,将像素点分为几个不同的簇,在代价聚合过程中,融合聚类分割信息、像素颜色信息和像素梯度信息,分析像素点间的差异性,自适应调整能量函数中的惩罚项,使视差突变处依然能保持特征。最后使用亚像素细化、左右一致性检查、中值滤波优化视差值,获得更精确的视差图。实验结果表明,改进的SGM算法生成的视差图相较于传统的SGM算法精确度更高,对不同森林场景的鲁棒性更好。Generating accurate disparity map is the basis of three-dimensional reconstruction and resource investigation under forest canopy by binocular vision.Due to the complex structure and many parallax discontinuities in the forest environment,the existing algorithms will have a large mismatch rate and unclear trunk edge contour when applied to forest scenes.Aiming at the above problems,this paper proposes an improved Semi-Global Matching(SGM)algorithm.Firstly,in the initial cost calculation stage,the Census central pixel value is replaced by the average value of four neighboring pixels to avoid the matching error caused by the sudden change of the central pixel value.Secondly,the binocular camera image is segmented by k-means clustering,and the pixels are divided into several different clusters.In the process of cost aggregation,the clustering segmentation information,pixel color information and pixel gradient information are fused,and the differences between pixels are analyzed,and the penalty term in the energy function is adaptively adjusted,so that the features of parallax mutation can still be maintained.Finally,sub-pixel thinning,left-right consistency checking and median filtering are used to optimize the disparity value and obtain a more accurate disparity map.Experimental results show that the disparity map generated by this algorithm is more accurate and more robust to different forest scenes than the traditional SGM algorithm.

关 键 词：森林场景 SGM 代价聚合 视差优化 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]