Leaf Area Estimation Method Based on Three-dimensional Point Cloud

In the study of 3D plant phenotyping, with the efficiency of plant phenotype improving, 3D reconstruction technology has become a novel direction to obtain blade morphology at present. However, the traditional method of reconstructing 3D model such as laser scanner, structured light image and binocular stereo vision system is expensive and complicated. A new method extracting plant leaf area was proposed based on point cloud obtained by using structure from motion. Smart phone was used to get images of plants. Based on these images, the three-dimensional point cloud of plants was reconstructed by using structure from motion algorithm. In order to restore the surface shape of the leaf, firstly, the noise of the leaf point cloud was removed by using threshold segmentation algorithm based on HSV color space. Secondly, the three-dimensional coordinate matrix of point cloud was classified by using K-means clustering algorithm to segment single leaf point cloud by classifying. And then, the surface mesh model of the leaf was reconstructed by using the ball pivoting algorithm. At last, the leaf area was obtained by calculating the mesh area. To evaluate the proposed method, it was compared with the conventional leaf area measurement method. The average error of the calculation result of the proposed method was 4.67% compared with the measured value by using the scanning method, and the average error was 6.05% compared with the measured value by using the leaf-shape paper weighing method. In addition, the method of calculating leaf area by extracting contour by Canny edge detection algorithm was compared with the proposed method. The results showed that the proposed method required low cost and high precision, and met the requirements of non-destructive and accurate determination of plant leaf area. © 2019, Chinese Society of Agricultural Machinery. All right reserved.