Abstract: This study focuses on the three-dimensional reconstruction method of maize seedlings based on RGB-D binocular vision, and optimizes some of the reconstruction parameters. Firstly, a fixed step angle surround image acquisition is performed on the target maize seedlings. According to the segmentation result of the target area in the RGB image, the depth data of the target area in the depth image is segmented. An improved mean filtering method is used to adaptively fill the depth data holes in the maize seedling area. Secondly, multi-angle point cloud registration and fusion are completed by first roughly and then finely processing the depth point cloud data of the maize seedlings at various angles. Finally, two voxel simplification methods are compared for their effectiveness in reducing and smoothing point clouds, achieving the reconstruction of the three-dimensional model of the maize seedlings. The efficiency and accuracy of maize seedlings modeling with different step angles are experimentally compared. The results show that using an octree filter achieves better simplification effects and the modeling error is minimized at a 60° step angle. The reconstructed model has a precision error of 4. 4 mm for the plant height and an average precision error of 0. 62 mm for the stem diameter, which meets the requirement for the three-dimensional reconstruction morphological measurement of maize seedlings.