Method for multi-view point cloud registration for whole aircraft based on spherical harmonic feature (SHF)

1. A method for multi-view point cloud registration for a whole aircraft based on a spherical harmonic feature (SHF), specifically comprising the following steps:

step S 1 : developing a measurement solution for a whole aircraft, and performing data preprocessing after point cloud data of the whole aircraft is obtained;

step S 2 : calculating a preprocessed point cloud based on step S 1 , performing local spherical projection on each point, and calculating a local SHF based on spherical harmonic transform; and

step S 3 : calculating a constraint correspondence between overlapping point cloud views based on the local SHF calculated in step S 2 , filtering the established constraint correspondence to remove an incorrect constraint correspondence, and based on an equation of an optimization graph, resolving transformation of point clouds from different viewing angles to implement multi-view point cloud registration.

2. The method for multi-view point cloud registration for a whole aircraft based on an SHF according to claim 1 , wherein the measurement solution in step S 1 specifically comprises deploying different viewing angles to scan the aircraft, and there are overlapping degrees of 40% to 60% between the point clouds from different viewing angles.

3. The method for multi-view point cloud registration for a whole aircraft based on an SHF according to claim 1 , wherein data preprocessing in step S 1 comprises:

step S 101 : performing downsampling on the point cloud data, making a sampled point set be S, randomly adding one point from the point clouds to S, finding, from a remaining point, a point with a largest sum of distances from all points in S, adding the point to S, and continuously updating S until a point quantity in S reaches a specified point quantity;

step S 102 : performing denoising on the point cloud data, traversing all points with a given radius r 1 and a threshold N by using point cloud denoising based on density, calculating a point quantity in the radius r 1 of each point which is denoted as M, and if M<N, deleting the point, otherwise, retaining the point; and

step S 103 : performing coarse registration on point clouds from multiple viewing angles, and roughly estimating gesture parameters of the point clouds by manually selecting at least three points at a same location in an overlapped region from different viewing angles, to implement coarse registration of multi-view point clouds.

4. The method for multi-view point cloud registration for a whole aircraft based on an SHF according to claim 1 , wherein step S 2 specifically comprises the following steps:

step S 201 : based on the point cloud data obtained in step S 1 , establishing a local sphere by using each point as a center of the sphere and by using r 2 as a radius, and constructing a Gaussian sphere feature by projecting a point in the local sphere onto a local sphere surface; and

step S 202 : performing spherical harmonic transform on the Gaussian sphere feature, converting the feature into a frequency domain, and resolving the feature into a form of a spherical harmonic function coefficient; and calculating energy of a spherical harmonic at different frequencies as the local SHF.

5. The method for multi-view point cloud registration for a whole aircraft based on an SHF according to claim 1 , wherein step S 3 specifically comprises the following steps:

step S 301 : judging a corresponding point based on the local SHF, and determining a constraint correspondence between corresponding points in overlapped point clouds in viewing angles;

step S 302 : performing filtering based on curvature of the point clouds, and when a difference between curvature of a pair of initially matched feature points is less than a specified threshold, judging that the points are corresponding points; and

step S 303 : establishing a target equation of a global optimization graph, applying local error diffusion to the entire graph, that is, evenly distributing an accumulated error to all views, and performing transformation on the point clouds based on a calculation result of the optimization graph, to implement multi-view point cloud registration, and complete splicing of point clouds of the whole aircraft.