Object Segmentation Ensuring Consistency Across Multi-Viewpoint Images.

Area of Science:

• Computer Vision
• Image Processing
• 3D Reconstruction

Background:

• Multi-view object segmentation is crucial for 3D scene understanding.
• Existing methods struggle with sparse, wide-baseline camera setups.
• Integrating color and depth information offers a promising solution.

Purpose of the Study:

• To develop a robust hybrid approach for multi-view object segmentation.
• To improve segmentation accuracy and inter-view consistency.
• To address challenges in sparse and wide-baseline capture environments.

Main Methods:

• Hybrid segmentation using color and depth from RGBD and sparse color cameras.
• Dense depth generation via Structure from Motion and Joint Bilateral Upsampling.
• Markov Random Field optimization on superpixel graphs with pixel-based boundary refinement.

Main Results:

• Achieved accurate object segmentation using combined color and depth data.
• Demonstrated effective inter-view geometric correspondence generation.
• Validated performance across various capture conditions and compared favorably to state-of-the-art methods.

Conclusions:

• The proposed hybrid method provides efficient and accurate multi-view object segmentation.
• It excels in challenging sparse, wide-baseline capture scenarios.
• This approach advances 3D scene reconstruction and analysis capabilities.