Color-guided depth recovery from RGB-D data using an adaptive autoregressive model

Area of Science:

• Computer Vision
• Image Processing

Background:

• Depth cameras often capture low-quality measurements.
• Accurate depth recovery is crucial for many applications.

Purpose of the Study:

• To propose an adaptive color-guided autoregressive (AR) model for high-quality depth recovery.
• To enhance depth map accuracy from low-quality measurements.

Main Methods:

• Formulating depth recovery as minimizing AR prediction errors under measurement consistency.
• Constructing AR predictors using local depth map correlations and nonlocal color image similarities.
• Analyzing method stability and designing a parameter adaptation scheme.

Main Results:

• The AR model effectively fits depth maps of generic scenes.
• Quantitative and qualitative evaluations demonstrate superiority over ten state-of-the-art methods.
• The method achieves stable and accurate depth recovery.

Conclusions:

• The proposed adaptive color-guided AR model significantly improves depth recovery quality.
• The method is versatile and effective for various depth degradations and sensors like Time-of-Flight and Kinect.