Consistent depth video segmentation using adaptive surface models.

Area of Science:

• Computer Vision
• Geometric Modeling
• 3-D Data Processing

Background:

• Accurate segmentation of 3-D point clouds is crucial for various applications.
• Existing methods often struggle with dynamic scenes and diverse data acquisition devices.
• Temporal coherence and traceability of segments are significant challenges in video-based segmentation.

Purpose of the Study:

• To develop a novel, adaptive approach for segmenting 3-D point clouds into geometric surfaces.
• To achieve temporally coherent and traceable surface segments in dynamic scenes.
• To validate the method's effectiveness across different range imaging technologies.

Main Methods:

• An iterative split-and-merge procedure is employed for segmentation.
• Adaptive mechanisms are incorporated for segment creation and removal.
• The approach utilizes adaptive surface models to adjust to changing input data.

Main Results:

• The proposed method successfully segmented 3-D point cloud videos acquired from structured-light and time-of-flight sensors.
• The segmentation produced stable, temporally coherent, and traceable surface segments.
• Quantitative evaluations using ground-truth data confirmed the approach's feasibility and accuracy.

Conclusions:

• The adaptive surface model offers a robust solution for 3-D point cloud segmentation.
• The iterative split-and-merge procedure with adaptive mechanisms enhances segmentation stability and temporal coherence.
• The method demonstrates broad applicability across various range imaging devices and data types.