| 摘要: |
| 【目的】对格木生长因子的双目测定方法进行研究，为格木等珍贵树种生长数据的无损、高效测定提供参考。【方法】利用MicaSense RedEdge多光谱相机获取格木图像，采用小波包变换与限制对比度自适应直方图均衡算法(contrast limited adaptive histogram equalization,CLAHE)对格木图像进行预处理；应用Matlab平台中的Stereo Camera Calibrator工具箱进行相机标定后，使用Bouguet算法对预处理的格木图像进行立体校正，并通过ADCensus和半全局匹配(SGM)算法进行立体匹配获得视差图像，然后根据三角测量原理计算格木树高、冠幅及地径。【结果】对由MicaSense RedEdge多光谱相机获取的格木图像，小波包变换可以有效去除图像中的强噪声，经小波包去噪后其平均信噪比(signal-to-noise ratio,SNR)达到了24.932 6，去噪效果显著优于小波阈值去噪及其他滤波方法；CLAHE均衡化后格木图像的灰度值分布概率更加均匀且噪声未被过度放大。通过相机标定获取左右相机的内外参数、平移旋转矩阵及畸变系数，得到的重投影误差为0.169 8个像素，实现了高精度标定；Bouguet立体校正实现了左右视图的行对准，将图像匹配的搜索维度由二维降至一维；相较于SGM算法，ADCensus算法立体匹配的视差图像更加稠密、平滑，对于格木生长因子的测算精度更高，基于ADCensus算法视差图像测算的树高、冠幅及地径与实际测量值的平均相对误差 (MRE) 分别为 2.108%，2.949%和5.938%，平均绝对误差 (MAE) 分别为1.429，1.346和0.030 cm，均方根误差 (RMSE)分别为1.525，1.390和0.031 cm。【结论】图像预处理可以减少强噪声并增强图像细节，提升重建的准确性。通过高精度标定、立体较正、立体匹配可以获取准确的视差信息，进而实现对格木生长因子的快速、精准测定。 |
| 关键词: 格木 生长因子 双目视觉 图像降噪 立体匹配 |
| DOI：10.13207/j.cnki.jnwafu.2025.09.007 |
| 分类号: |
| 基金项目:中央级公益性科研院所基本科研业务费专项(CAFYBB2021ZB002)；国家自然科学基金面上项目(32071761) |
|
| Study on binocular measurement of growth factors of Erythrophleum fordii |
|
WANG Peng, WANG Xuefeng
|
|
Key Laboratory of Forest Management and Growth Simulation,National Forestry and Grassland Administration，Institute of Forest Resource Information Techniques,Chinese Academy of Forestry,Beijing 100091,China
|
| Abstract: |
| 【Objective】The binocular determination method of growth factor of Erythrophleum fordii was studied to provide a reference for non-destructive and efficient determination of growth data of precious tree species such as E.fordii.【Method】The MicaSense RedEdge multispectral camera was used to acquire the images of the E.fordii.The wavelet packet transform and the contrast limited adaptive histogram equalization (CLAHE) algorithm were used to preprocess the images of the E.fordii.The camera was calibrated using the Stereo Camera Calibrator toolbox in the Matlab platform.The Bouguet algorithm performed stereo correction on the preprocessed E.fordii images.The ADCensus and Semi Global Matching (SGM) algorithms were used for stereo matching to obtain the disparity images.The height,crown width,and ground diameter of the E.fordii were then calculated based on the triangulation principle.【Result】For the E.fordii images acquired by the MicaSense RedEdge multispectral camera,wavelet packet transform was used to effectively remove strong noise from the image.After wavelet packet denoising process,the average signal-to-noise ratio (SNR) reached 24.932 6,indicating a significantly better denoising effect than wavelet threshold denoising method and other filtering methods.After CLAHE equalization,the grayscale value distribution probability of the E.fordii image was more evenly distributed and the noise was not over-amplified.Through camera calibration,the internal and external parameters,translation and rotation matrices,and distortion coefficients of the left and right cameras were obtained,and the reprojection error was 0.169 8 pixels,achieving high-precision calibration.Bouguet stereo correction achieved row alignment of the left and right views,and the image matching was transformed from two dimension search to one dimension search.Compared with the SGM algorithm,the disparity image of the stereo matching of the ADCensus algorithm was denser and smoother,and the measurement accuracy of the growth factor of the E.fordii was higher.The tree height,crown width,and ground diameter calculated based on the disparity image of the ADCensus algorithm showed a mean relative error (MRE) of 2.108%,2.949% and 5.938% compared with the measured values,respectively.The mean absolute error (MAE) was 1.429,1.346 and 0.030 cm,respectively,and the root mean square error (RMSE) was 1.525,1.390 and 0.031 cm,respectively.【Conclusion】Image preprocessing can reduce strong noise,enhance image details,and improve the accuracy of reconstruction.Accurate parallax information can be obtained through high-precision calibration,epipolar correction,and stereo-matching,to realize the rapid and precise determination of the growth factors of E.fordii. |
| Key words: Erythrophleum fordii growth factor binocular vision image denoising stereo matching |
