On Stereo Confidence Measures for Global Methods: Evaluation, New Model and Integration into Occupancy Grids

Area of Science:

• Computer Vision
• 3D Reconstruction
• Machine Learning

Background:

• Stereo confidence measures are crucial for accurate 3D reconstruction.
• Existing methods have limitations in evaluating confidence across the entire disparity range.

Purpose of the Study:

• To evaluate and compare existing stereo confidence models.
• To introduce a novel Histogram Sensor Model (HSM) for improved stereo confidence.
• To develop a systematic parameter estimation method for parametric models.

Main Methods:

• Comparison of multiple stereo confidence models across different cost functions and window sizes.
• Introduction and evaluation of the Histogram Sensor Model (HSM).
• Systematic parameter estimation for parametric stereo confidence models.
• Application in a 3D reconstruction framework using occupancy grids.

Main Results:

• The proposed Histogram Sensor Model (HSM) demonstrates superior overall performance.
• The new parameter estimation method yields better results than previous approaches.
• Evaluations were conducted on both indoor and outdoor datasets for comprehensive analysis.

Conclusions:

• The Histogram Sensor Model (HSM) is a highly effective stereo confidence measure.
• Systematic parameter estimation significantly enhances the performance of parametric models.
• The findings are relevant for both winner-take-all stereo and global 3D reconstruction applications.