Real-Time Plane Detection with Consistency from Point Cloud Sequences.

Area of Science:

• Computer Vision
• Robotics
• 3D Data Processing

Background:

• Real-time consistent plane detection (RCPD) is crucial for advanced computer vision and robotic applications.
• Existing methods struggle with long processing times, imprecise plane detection, and inconsistent plane labeling across image sequences.
• These limitations hinder the reliable application of plane detection in dynamic environments.

Purpose of the Study:

• To develop a novel superpixel-based approach for real-time consistent plane detection (RCPD).
• To enhance the accuracy and consistency of plane detection in structured 3D point cloud sequences.
• To address the limitations of existing methods regarding speed, precision, and temporal consistency.

Main Methods:

• Proposed a real-time plane detection algorithm for extracting planes from raw 3D point clouds acquired by depth sensors.
• Implemented a superpixel-based segmentation technique to ensure precise matching of detected planes to their actual boundaries.
• Developed a robust strategy for recovering lost planes by leveraging contextual correspondence information from adjacent frames.

Main Results:

• The superpixel-based method achieves real-time performance in plane detection from 3D point clouds.
• The approach demonstrates improved accuracy in segmenting planes, aligning detected boundaries with actual object edges.
• The strategy effectively recovers missing planes, maintaining consistent plane identification across frames.

Conclusions:

• The novel superpixel-based real-time plane detection method significantly outperforms state-of-the-art techniques.
• The approach offers superior efficiency and accuracy for plane detection in structured 3D point cloud sequences.
• This work provides a robust solution for consistent plane detection, advancing computer vision and robotics.