基于消费级双目相机的立木因子测量方法

  目的  随着林业信息化的快速发展，机器视觉测量技术广泛应用于林业领域。针对传统立木因子测量方法成本较高、携带不便、操作复杂等问题，提出消费级双目相机与机器视觉技术相结合的立木因子无接触测量方法。  方法  首先使用消费级USB 3.0双目相机采集立木图像，通过改进的SGM算法生成高质量视差图；再根据三角原理转化为深度图，进而获取立木三维点云；基于空间密度聚类和混合滤波三维点云去噪方法快速准确去除聚集、离散的噪声点，再进行方向矫正和点云分割；最后，利用最值遍历法和椭圆拟合法实现树高、胸径的无接触测量。  结果  树高、胸径的相对测量误差分别小于2.219%、5.620%，测量值与真实值的相关系数R2分别为0.978、0.995，均方根误差分别为0.047 m、0.249 cm。  结论  本方法易操作、成本较低，同时具有较高的测量精度，能够满足无接触测量的需求。图5表2参27

Measuring method of tree attributes based on consumer-grade binocular camera

  Objective  With machine vision measurement technology being widely used in the field of forestry as a result of the rapid development of forestry informatization, this study is aimed to propose a contact less tree attribute measuring method combining consumer binocular-grade camera and machine vision technology to replace the traditional one which is featured with high cost, low mobility and complicated operation.   Method  Firstly, a consumer-grade USB 3.0 binocular camera was used to capture images of tree before a high-quality parallax image was generated by an improved SGM algorithm. Then it was transformed into a depth image in accordance with the triangulation principle so as to obtain a 3D point cloud. Next, the three-dimensional point cloud denoising method based on spatial density clustering and hybrid filtering was employed to remove the aggregated and discrete noise points quickly and accurately after which orientation correction and point cloud segmentation were performed. Finally, the most-valued traversal method and ellipse fitting method were used to achieve contactless measurement of tree height and DBH (diameter at breast height).  Result  The relative measurement errors of tree height and DBH were less than 2.219% and 5.620%, with the correlation coefficients being 0.918 and 0.995, whereas the root mean square errors being 0.047 m and 0.249 cm respectively.   Conclusion  The proposed method in this paper, featured as convenient with low cost and high precision, can meet the requirements of contactless measurement. [Ch, 5 fig. 2 tab. 27 ref.]