Depth-color fusion strategy for 3-D scene modeling with Kinect

Area of Science:

• Computer Vision
• Human-Computer Interaction
• 3-D Modeling

Background:

• Microsoft Kinect depth data suffers from noise, affecting accuracy in human-computer interaction.
• Improving depth data quality is crucial for reliable controller-free gaming and 3-D scene reconstruction.

Purpose of the Study:

• To present a depth-color fusion strategy for accurate 3-D modeling of indoor scenes using Kinect.
• To address noise-related problems in Kinect depth data, including distance-dependent inaccuracies, spatial noise, and temporal fluctuations.

Main Methods:

• Iterative building of accurate depth and color background models.
• Utilizing an adaptive joint-bilateral filter to fuse depth and color information.
• Analyzing edge-uncertainty maps and detected foreground regions for filtering.

Main Results:

• Significant reduction in distance-dependent depth maps, spatial noise, and temporal fluctuations.
• Refined object depth boundaries and interpolation of non-measured depth pixels.
• Generation of robust depth/color background models and accurate moving object silhouettes.

Conclusions:

• The proposed depth-color fusion strategy effectively enhances Kinect depth data quality.
• Improved data accuracy leads to more reliable 3-D modeling and human-computer interaction.
• The method broadens the applicability of Kinect for advanced scene understanding and control.