A three-dimensional point cloud RANSAC pose estimation method and system fusing two-dimensional radius prior

By integrating a 2D radius prior with a 3D point cloud RANSAC pose estimation method, the inaccuracy and stability problems of 3D pose estimation for pineapple fruits are solved, achieving high precision and high stability in pineapple fruit harvesting, which is suitable for intelligent pineapple harvesting.

CN122115559APending Publication Date: 2026-05-29SOUTH CHINA AGRICULTURAL UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN Β· China
Patent Type
Applications(China)
Current Assignee / Owner
SOUTH CHINA AGRICULTURAL UNIVERSITY
Filing Date
2026-01-26
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing 3D pose estimation technology for pineapple fruits suffers from inaccurate 3D pose recovery, poor adaptability, and insufficient model stability, failing to meet the accuracy and stability requirements of intelligent pineapple harvesting.

Method used

A RANSAC pose estimation method for 3D point clouds with fusion of 2D radius prior is adopted. A binary mask image is generated by an instance segmentation network and a color point cloud is generated by combining the depth image. After noise reduction, the radius prior interval constraint RANSAC algorithm is introduced to select the optimal cylindrical model and extract the 3D pose information.

Benefits of technology

This method enables precise construction of pineapple fruit pose, improves harvesting success rate, reduces fruit damage probability, and enhances the stability and accuracy of 3D pose detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122115559A_ABST
    Figure CN122115559A_ABST
Patent Text Reader

Abstract

The application discloses a three-dimensional point cloud RANSAC pose estimation method and system fusing a two-dimensional radius prior, and the method is as follows: inputting a collected RGB image into an instance segmentation network model after pretreatment to obtain a binary mask image, combining a depth image to generate an original pineapple fruit color point cloud and performing noise reduction processing on the original pineapple fruit color point cloud; performing fitting processing on the binary mask image to obtain core parameters of a rectangle; determining a representative depth value in combination with the depth image, converting pixel length of the rectangle into physical length to preliminarily estimate a fruit radius, introducing a scale calibration coefficient to correct the radius prior interval; taking the radius prior interval as a constraint to improve the RANSAC algorithm, randomly sampling the pineapple fruit color point cloud to generate a candidate cylindrical model and eliminate invalid models; performing consistency inspection on the candidate cylindrical model meeting the radius prior interval constraint, selecting an optimal cylindrical model, and extracting a direction vector of the cylindrical axis and an arbitrary point on the axis as three-dimensional pose information of the pineapple fruit.
Need to check novelty before this filing date? Find Prior Art