Reliable Disparity Estimation Using Multiocular Vision with Adjustable Baseline

Area of Science:

• Computer Vision
• 3D Reconstruction
• Stereo Vision

Background:

• Accurate 3D information estimation is crucial for computer vision.
• Binocular stereo vision's reliability depends on camera baseline, posing trade-offs between resolution and matching errors.

Purpose of the Study:

• To develop a reliable method for disparity estimation using progressive baseline increases in multiocular vision.
• To improve accuracy and reduce matching errors in 3D scene characterization.

Main Methods:

• Introduced a robust multiocular image rectification method satisfying epipolar constraints.
• Estimated dense disparity maps via stereo matching with progressively increasing baselines.
• Iteratively refined disparity maps to minimize matching errors and error propagation.

Main Results:

• Achieved a 25% improvement in rectification error for binocular and 80% for multiocular images.
• Increased disparity estimation accuracy by 20% for multiocular images compared to existing methods.
• Demonstrated superior performance on multiocular image datasets and physical scenes.

Conclusions:

• The proposed progressive baseline method enhances 3D information estimation accuracy.
• Enables precise scene characterization and spatial point computation.
• Offers a reliable solution for challenging multiocular vision tasks.