NCC-RANSAC: a fast plane extraction method for 3-D range data segmentation.

Area of Science:

• Computer Vision
• Geometric Modeling
• 3D Reconstruction

Background:

• Standard Random Sample Consensus (RANSAC) plane extraction can over-extract planes or fail on complex scenes like stairways.
• Connected inlier patches in multistep scenes pose challenges for existing CC-RANSAC methods, leading to inaccurate plane fitting.

Purpose of the Study:

• To introduce a novel Normal-Coherence RANSAC (NCC-RANSAC) method for robust plane extraction.
• To address limitations of existing RANSAC-based methods in handling connected inlier patches in complex 3D scenes.

Main Methods:

• The NCC-RANSAC algorithm incorporates a normal coherence check to identify and remove data points with contradictory normal directions within inlier patches.
• This process segments connected patches into distinct candidate planes, which are then recursively clustered and grown.
• A probabilistic model is developed to predict the algorithm's success probability.

Main Results:

• NCC-RANSAC successfully separates connected inlier patches, enabling accurate extraction of individual planes in complex scenes.
• Experimental validation using a 3-D time-of-flight camera (SwissRanger SR4000) demonstrated superior accuracy compared to existing RANSAC variants.
• The proposed method achieves higher accuracy with reduced computational time.

Conclusions:

• NCC-RANSAC offers a significant advancement in plane extraction for 3D data, particularly in challenging scenarios with connected surfaces.
• The method provides a more reliable and efficient solution for 3D scene understanding and reconstruction.
• Further research could explore the probabilistic model for real-time applications.