Three-dimensional Point Cloud Registration Method for Soil Surface Based on Kinect Camera

In order to establish a better three-dimensional point cloud morphological structure model of soil surface, a Kinect camera was used to obtain color images and depth images of the soil surface. For the traditional nearest point iterative algorithm in the point cloud registration, the space position requirements are more stringent, thus the method of initial registration of point cloud was proposed. Firstly, it was necessary to remove the useless background information and noise of the depth image of the acquired soil surface, and then initial registration and precise registration of the three-dimensional point cloud were performed. In the initial registration process, the point cloud information of the acquired soil surface was normalized and aligned to the radial feature key point search to obtain representative and relatively uniform point cloud key points, and then the fast point feature value histogram method was used to extract the eigenvalues of the key points, and finally the random sampling consistency algorithm was used to purify the mapping relationship, thereby completing the initial registration of the point cloud. Finally, the nearest point iteration algorithm was used to accurately register the three-dimensional point cloud on the soil surface. The registration time of the traditional nearest point iterative algorithm was 58.2 s, the registration error was 3.80 cm, the improved method registration time was 124.8 s, and the registration error was 0.89 cm. Compared with the traditional nearest point iterative algorithm, the registration time of the improved method was extended by 66.6 s, but the registration error was reduced by 2.91 cm. The results showed that the method was simple, easy to handle, and low in cost, and can realize three-dimensional reconstruction of the soil surface. © 2019, Chinese Society of Agricultural Machinery. All right reserved.