Robust multi-model fitting method based on structural decision graph

Abstract

The invention provides a robust multi-model fitting method based on a structural decision graph, and relates to robust multi-model fitting. The method includes following steps: performing random sampling on input sample data to generate a lot of hypotheses; calculating the weight score of each generated hypothesis based on kernel density estimation and interior point scale; ranking all the hypotheses according to the weight scores; calculating the shortest arrival distance of each ranked hypothesis based on the consensus set and the Pearson production-moment correlation coefficient; constructing the structural decision graph according to the weight scores and the shortest arrival distance; determining structural prototypes corresponding to all structures on the structural decision graph, and calculating the number of the structures; and differentiating interior points and anomalous points according to the structural prototypes, outputting a model parameter corresponding to each structure, and accomplishing robust multi-model fitting based on the structural decision graph. According to the method, the structural prototype is selected by employing hypothetical consensus information, the filtering or clustering process is not involved, and problems of possibility of deletion of representative hypotheses and negligence of small structures is solved.

Description

technical field [0001] The invention relates to robust multi-model fitting, in particular to a robust multi-model fitting method based on a structural decision graph. Background technique [0002] In a range of computer vision applications such as 3D planar reconstruction and motion segmentation, observation data from multi-model distributions usually contain a large number of outliers. How to extract multi-structural information from these outlier-containing data is a major challenge for these applications. The tasks of robust multiple model fitting including outliers include: 1) accurately estimating the number of structures present in the data; 2) accurately recovering the model parameters corresponding to these structures. [0003] In the field of computer vision, a large number of robust multi-model fitting algorithms have been proposed to deal with data containing outliers. Among these proposed algorithms, some algorithms achieve the task of fitting and segmenting mu...

AI Technical Summary

Problems solved by technology

However, a possible problem is that when the structure is unbalanced in the data, the filtering step may remove some smaller structures containing few inliers

Images