Surface Reconstruction via Fusing Sparse-Sequence of Depth Images.

Area of Science:

• Computer Vision
• 3D Reconstruction
• Robotics

Background:

• Handheld scanning with depth cameras offers low-cost 3D model generation.
• Existing dense fusion methods suffer from data redundancy and jittering frames, leading to poor model quality and scan failures.
• Camera tracking loss is a significant challenge in real-time handheld scanning.

Purpose of the Study:

• To propose a novel sparse-sequence fusion (SSF) algorithm for efficient and robust handheld 3D scanning.
• To reduce data redundancy and mitigate the impact of jittering frames in depth image sequences.
• To enhance the quality and reliability of 3D models reconstructed from commodity depth cameras.

Main Methods:

• The sparse-sequence fusion (SSF) algorithm analyzes camera motion using extracted measurements.
• A supporting subset of depth images is progressively constructed to decrease redundancy and jitter.
• A refinement process eliminates noise and recovers geometric features from selected depth images.
• Refined depth images are integrated into the truncated signed distance field (TSDF) for final fusion.

Main Results:

• The SSF algorithm effectively reduces data redundancy compared to dense fusion methods.
• Mitigation of jittering frames leads to improved feature clarity and reduced blurring in reconstructed models.
• Experimental comparisons demonstrate the feasibility and validity of SSF for handheld scanning.
• The method successfully recovers geometric features and eliminates noise from depth images.

Conclusions:

• The proposed sparse-sequence fusion (SSF) algorithm significantly enhances handheld 3D scanning using commodity depth cameras.
• SSF addresses key challenges of data redundancy and frame jitter, leading to higher quality 3D reconstructions.
• This approach offers a more robust and efficient solution for low-cost 3D modeling applications.