Space-time stereo analysis combining local structure and modulation features in the monogenic wavelet domain.

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• Estimating time-varying disparity maps is crucial for stereo vision applications.
• Existing methods often struggle with variations in lighting, scale, and rotation.
• Intensity-based cost functions can be sensitive to these image variations.

Purpose of the Study:

• To propose a novel multimodal cost function for robust disparity map estimation.
• To enhance the accuracy and reliability of stereo vision systems.
• To address the limitations of traditional cost functions in handling image variations.

Main Methods:

• A multimodal cost function integrating local structure and modulation information from the monogenic wavelet transform.
• Utilizing constraints on local orientation, phase, and amplitude congruencies.
• Employing weighted coefficients adapted to local image features for invariance to level shift, scaling, rotation, and lighting.

Main Results:

• The proposed cost function significantly improves robustness in estimating time-varying disparity maps.
• Experimental results on synthetic and natural stereo sequences outperform the standard sum of squared difference (SSD) cost function.
• The method shows insensitivity to common image distortions.

Conclusions:

• The developed multimodal cost function offers a more robust and reliable solution for stereo vision disparity estimation.
• This approach enhances the performance of stereo vision systems in challenging and dynamic environments.
• The findings pave the way for improved 3D reconstruction and scene understanding.