257 / 2024-03-14 15:59:46

3D reconstruction of mine environment based on a two-step point cloud registration algorithm

Machine vision; Point cloud registration; 3D reconstruction;

4、智能化装备技术 > 6、装备智能感知技术

A two-step point cloud registration algorithm based on improved sampling consistency and iterative nearest neighbors is proposed to meet the daily work needs of coal mine production robots and the rapid rescue needs of post disaster rescue robots. This algorithm helps to obtain a three-dimensional environment reconstruction model of restricted scenes in coal mines.

Firstly, the depth maps acquired by the Kinect-2 camera are image-aligned with the colour maps to generate a 3D point cloud and perform pre-processing operations including point cloud refinement, outlier removal, and creation of 3D topology; Then, in response to the unstable geometric features of randomly sampled points and the high computational cost of Fast Point Feature Histograms (FPFH) features in the sampling consistency algorithm, the point cloud data is simplified using average curvature, combined with two constraints of distance threshold and point cloud average density, to generate a feature point set and matching point set with richer geometric features. The transformation relationship between the corresponding points in the point set is calculated, and the initial transformation matrix of the point cloud is obtained through point cloud coarse registration; Finally, to address the issue of inaccurate initial values and the risk of falling into local optima, an iterative nearest neighbor point cloud registration algorithm combined with the initial transformation matrix is used for precise registration.

Under the unstructured simulation environment of the mine and the simulation environment of the mine tunnel, the proposed algorithm has a high degree of realism of the reconstructed model, and the accuracy of the point cloud alignment is better than that of the mainstream model, so the feasibility of the algorithm is verified.