Abstract:The 3D structure reconstruction of plants is one of the important methods to realize the no-destructive measurement of plant morphological structure. Consumer depth cameras are widely used to acquire 3D spatial information because of their advantages of noncontact, quick acquisition of data and low cost. In 3D reconstruction, feature points were usually extracted from the point cloud, then paired with feature points, and the registration matrix T was calculated. However, it was difficult to extract the characteristic points of plants because of their slender body. According to the need of plant phenotypic analysis for 3D reconstruction and aiming at the problem that it was difficult to extract plant feature points which resulted in a bad 3D reconstruction, a 3D reconstruction method based on depth camera was proposed. Firstly, the depth camera was calibrated internally and the depth value distortion was corrected by which the accurate depth value can be obtained. Then the relative position of the camera and the turntable were fixed, and the matrix T corresponding to a fixed angle θ of turntable was accurately calculated. The matrix T was calculated in the coordinate space of the current depth camera with the help of a chessboard. After obtaining the matrix T, the turntable was rotated at an interval of θ to get a series of point clouds which were registered as a whole one with the help of transformation matrix T to complete 3D reconstruction. By comparing the reconstruction results of commercial software Skanect, this method needs only one registration, and has higher reductive degree, better efficiency and robustness, which meets the needs of plant morphology measurement.
