Abstract: To address the problems  of incomplete characterization of phenotypic information and low accuracy stemming from the current monocular image measurement of fluecured tobacco surface area, a novel method based on an improved SemiGlobal Block Matching SGBM algorithm was proposed for measuring the frontal surface area of fluecured tobacco. Initially, the method involved obtaining a Disparity Map through the application of camera calibration parameters and the SGBM algorithm incorporating the Census transform fusion BT algorithm. Subsequently, hole filling in the Disparity Map was achieved by using a guided filter to derive spatial threedimensional information for each point in the image. Following this, point cloud preprocessing was conducted through statistical filtering and point cloud segmentation. Finally, the Poisson algorithm was employed for threedimensional surface reconstruction, facilitating the calculation of the reconstructed front surface area to determine the tobacco leaf area. A comparative analysis was then conducted between the frontal surface area of fluecured tobacco calculated by using the grid method, a twodimensional image method, and the proposed method. Results indicate that when the frontal surface area calculated by the grid method is considered as the reference value, the surface area of the tobacco leaf measured by the proposed method exhibits closer proximity to the reference value than that obtained by the twodimensional image method, with an average accuracy increase of 30.7%.

Key words: surface area of fluecured tobacco, binocular vision, 3D point cloud, 3D reconstruction

摘要： 为解决当前单目图像测定烤烟表面积带来的表型信息表征不全面、精度低等问题，提出基于改进SGBM算法的烤烟正面表面积测量方法。首先，通过相机标定参数并结合Census变换融合BT算法的SGBM算法得出视差图，利用引导滤波对视差图孔洞进行填充，得到图像中各点的空间三维信息；其次，使用统计滤波、点云分割进行点云预处理；最后，采用泊松算法进行三维表面重建，并计算重建的正面表面积得到烟叶面积。对比网格法、二维图像法、本方法计算的烤烟正面表面积，结果表明，以网格法计算的正面表面积作为真实值，所提方法测量的烟叶表面积与二维图像法相比更接近真实值，平均准确率提高30.7%。

关键词: 烤烟表面积, 双目视觉, 三维点云, 三维重建