Automatic Super-Surface Removal in Complex 3D Indoor Environments Using Iterative Region-Based RANSAC.

Area of Science:

• Computer Vision
• 3D Data Processing
• Robotics

Background:

• Removing bounding surfaces (super-surfaces) like walls and floors from 3D point clouds is crucial for applications such as object recognition and human tracking.
• Existing methods like Random Sample Consensus (RANSAC) struggle with cluttered scenes, leading to incorrect foreground object removal due to random sampling and limited scene knowledge.

Purpose of the Study:

• To introduce a novel super-surface removal technique for segmenting and removing bounding surfaces from 3D point clouds in complex indoor environments.
• To address the limitations of existing methods in accurately distinguishing and removing background surfaces without affecting foreground objects.

Main Methods:

• The proposed method preprocesses unorganized point cloud data from commercial depth sensors.
• The point cloud is divided into four overlapping local regions.
• An iterative surface removal approach is applied to each region to segment and remove bounding surfaces.

Main Results:

• The novel technique demonstrates robust performance in super-surface removal and size reduction for complex 3D indoor environments.
• Evaluation using specificity, precision, recall, and F1 score on generated datasets showed performance metrics between 90% and 99%.

Conclusions:

• The developed method effectively removes bounding surfaces from 3D point clouds in challenging indoor environments.
• The approach offers improved accuracy and robustness compared to traditional methods, preserving essential foreground data for computer vision tasks.