CN-116012590-B - Basic matrix fitting method based on residual ordering and sparse limitation

Abstract

The invention provides a basic matrix fitting method based on residual ordering and sparse limitation, which generates an effective distance matrix for a data set containing a plurality of model examples and high-proportion outliers, and specifically comprises the following steps of S1, setting input data The method comprises the steps of selecting effective data by using sampling weight and information theory principle through an improved algorithm AGS, selecting a data subset from the effective data to generate M model hypotheses, and calculating an identification distance matrix by using the generated M model hypotheses, wherein the number of model instances is S and the sampling times is M Step S4, using information theory sum Calculating a sparse distance matrix Step S5, in Performing spectral clustering using parameter S to obtain The method can effectively combine the advantages of residual sequencing and sparse limitation, and can efficiently generate a more accurate distance matrix for data containing a plurality of model examples and high-proportion outliers.

Inventors

• LAI TAOTAO
• LI ZUOYONG
• MING RUI

Assignees

• 闽江学院

Dates

Publication Date

20260505

Application Date

20230222

Claims (2)

1. 1. The basic matrix fitting method based on residual ordering and sparse limitation is characterized by comprising the following steps of: the fitting method generates an effective distance matrix for a data set containing a plurality of model examples and high-proportion outliers, and specifically comprises the following steps of: Step S1, setting input data The number S of model instances and the sampling times M; S2, selecting effective data by using sampling weight and information theory principle through an improved algorithm AGS, and then selecting a data subset from the effective data to generate M model hypotheses; step S3, calculating an identification distance matrix by using the generated M model hypotheses ; Step S4, using information theory sum Calculating a sparse distance matrix ; Step S5, in Performing spectral clustering using parameter S to obtain Class marks of (2); the step S3 specifically comprises the following steps: Step S31 assumption for each model Calculate input data And (3) with Residual of (2) ; Step S32, estimating by using the scale estimator MSSE Is of the inner point scale of (2) ; Step S33, the elements of matrix H of the model are calculated as follows Formula 1 Step S34, calculating an identification distance matrix ; The calculation of the sparse distance matrix in step S4 specifically includes the following steps: Step S41 for sparse distance matrix Line i element of (2) Calculating the distance between the maximum element and the jth element as ; Step S42. Probability of the j-th element is calculated as ; Step S43, calculation of the threshold value of the ith line as ; Step S44 of setting the value smaller than The element of (2) is set to 0, otherwise its value is unchanged, the formula is as follows A formula (II); step S45, matrix the distance After performing the four steps described above on each row of the matrix obtained Not symmetrical, but the required distance matrix is symmetrical, by Obtaining a symmetrical matrix; The performing spectral clustering in the step S5 specifically includes the following steps: Step S51, calculating degree matrix ; Step S52, calculating Laplace matrix ; Step S53, calculate Sequencing the characteristic values from small to large, taking the former S characteristic values, and calculating the characteristic vectors of the former S characteristic values ; Step S54, composing the S feature vectors into a matrix ; Step S55, order Is that Is used to determine the vector of the i-th row of (c), wherein i=1, 2,. -%, n; Step S56 of Unitized such that ; Step S57, using k-means algorithm to get new sample points Clustering into clusters The method comprises the steps of randomly selecting S objects as initial clustering centers, then executing the following processes, The process (1) calculates the distance between each object and each cluster center, and distributes each object to the cluster corresponding to the cluster center nearest to the object; updating each cluster center to be an average value of all points of the cluster; Repeating processes (1) and (2) until no objects are reassigned to different cluster centers; the fitting method is used in the field of computer vision or artificial intelligence, and when the fitting method is used for dividing two-dimensional moving objects in a scene graph, a distance matrix is generated by scene graph data containing a plurality of two-dimensional moving objects, and the fitting method is carried out in the step S5 The data belonging to the two-dimensional moving object in the image is divided into points belonging to different moving objects.
2. 2. The method for fitting a basis matrix based on residual ordering and sparse constraint according to claim 1, wherein the fitting method is used for a robust model for feature-based image registration, pose estimation, multi-objective tracking, PPS signal estimation, FM signal estimation, or 3D motion segmentation.

Description

Basic matrix fitting method based on residual ordering and sparse limitation Technical Field The invention relates to the technical field of computer vision, in particular to a basic matrix fitting method based on residual error ordering and sparse limitation. Background Robust model estimation is an important research area for computer vision and artificial intelligence. It has been widely used in computer vision and artificial intelligence such as feature-based image registration, pose estimation, multi-target tracking, PPS and FM signal estimation, and 3D motion segmentation. Robust model estimation refers to estimating model hypotheses for all model instances (i.e., structures) from input data given a geometric model (e.g., a basis matrix), and then separating outliers and inliers belonging to different structures into different groups based on the estimated model hypotheses. The method comprises the steps of firstly constructing a distance matrix based on a spectral clustering algorithm, then performing spectral clustering on the distance matrix, and dividing data points into different groups. Such algorithms exhibit good performance over several common test data sets, such as ADELAIDERMF and Hopkins 155. There are mainly two methods for constructing the distance matrix, namely a method based on model estimation and a method based on non-model estimation. The former type is more popular because it is less affected by lost data and is robust to outliers. In order to generate accurate model hypotheses, the sampling algorithm of the model estimation method needs to sample at least a clean minimum subset. The clean minimum subset includes a minimum number of data points that belong to the same structure used to estimate the geometric model. Sampling algorithms can be divided into two types, random sampling algorithms and guided sampling algorithms. Random sampling was first used in RANSAC, which is still widely used in many model estimation methods due to its simplicity. However, to sample a clean minimum subset using random sampling, the number of samples required grows exponentially as the dimension of the geometric model increases. To alleviate the drawbacks of random sampling, a number of guided data sampling algorithms have been proposed. Rather than randomly sampling each data point with an equal sampling weight value, the guided sampling algorithm assigns each data point a sampling weight calculated from various information (e.g., matching score or preference analysis). The guided sampling algorithms can be divided into four classes, spatial proximity based, matching score based, greedy search based, and preference analysis based algorithms. Residual ordering based sampling algorithms are one of the most potential guided sampling algorithms, as such algorithms are good at finding interior points that belong to the same model instance (i.e., structure). The preference analysis based algorithm calculates sampling weights from the residual index of the preference analysis. Early proposed residual-based sampling algorithms required the calculation of the sampling weights of all input data (w-1) times to sample a subset of data with w data points. In order to improve the calculation efficiency, an algorithm has been proposed that calculates sampling weights of portions other than all input data by using only local constraints. However, to sample the smallest subset, the algorithm still needs to calculate the sampling weights (w-1) times. In order to further increase the computational efficiency, an improved algorithm AGS is then proposed, which selects valid data by using the sampling weights and the principle of information theory, and then selects a subset of data from the valid data. AGS calculates sampling weights only once for a subset of the sampled data. Experimental results indicate that AGS is superior to a variety of guided sampling algorithms. The present invention proposes a solution to the above-mentioned problems by first, the present invention fully analyzes and studies the latest literature for robust model estimation, including data sampling and model selection algorithms. Secondly, the invention generates a potential model hypothesis by selecting an effective sampling algorithm based on preference analysis to calculate a distance matrix with discrimination capability, thereby providing an effective robust model estimation method. Then, the invention uses the principle of information theory to obtain a sparse distance matrix with more discrimination capability. And finally, performing spectral clustering on the sparse distance matrix to accurately divide the data. The sampling method provided by the invention can be used in robust model fitting, so that the sampling method is applied to feature-based image registration, attitude estimation, multi-target tracking, PPS and FM signal estimation and 3D motion segmentation. Disclosure of Invention The invention provides a basic matrix fitting