A Fast Point Cloud Recognition Algorithm Based on Keypoint Pair Feature.

Area of Science:

• Computer Vision
• Robotics
• 3D Reconstruction

Background:

• Point Pair Feature (PPF) algorithms offer robust matching for point cloud recognition, even with occlusion.
• However, PPF algorithms can be inefficient due to superfluous feature point pairs in global descriptions.

Purpose of the Study:

• To propose an improved 6D pose estimation method using Keypoint Pair Feature (K-PPF) voting.
• To enhance the efficiency and accuracy of point cloud recognition by reducing redundant feature pairs.

Main Methods:

• The Keypoint Pair Feature (K-PPF) algorithm is developed, building upon the PPF algorithm.
• Keypoints are extracted using a combination of curvature-adaptive and grid ISS, with angle-adaptive judgment.
• The method employs a voting strategy for 6D pose estimation.

Main Results:

• The K-PPF algorithm demonstrates improved recognition efficiency and robustness compared to the original PPF algorithm.
• Experimental results show a reduction in redundant point pairs.
• The proposed method achieved over 12.5% improvement in recall rate compared to FPFH, CSHOT, SHOT, and SI algorithms.

Conclusions:

• The K-PPF voting method offers a more efficient and robust solution for 6D pose estimation in point cloud recognition.
• The keypoint extraction strategy effectively improves feature matching accuracy.
• This approach is particularly effective in scenes with varying levels of occlusion and complexity.