Superpixel Cost Volume Excitation for Stereo Matching

Shanglong Liu,Lin Qi,Junyu Dong,Wenxiang Gu,Liyi Xu

DOI: https://doi.org/10.1007/978-981-97-8508-7_2

2024-11-20

Abstract:In this work, we concentrate on exciting the intrinsic local consistency of stereo matching through the incorporation of superpixel soft constraints, with the objective of mitigating inaccuracies at the boundaries of predicted disparity maps. Our approach capitalizes on the observation that neighboring pixels are predisposed to belong to the same object and exhibit closely similar intensities within the probability volume of superpixels. By incorporating this insight, our method encourages the network to generate consistent probability distributions of disparity within each superpixel, aiming to improve the overall accuracy and coherence of predicted disparity maps. Experimental evalua tions on widely-used datasets validate the efficacy of our proposed approach, demonstrating its ability to assist cost volume-based matching networks in restoring competitive performance.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper attempts to solve the problem of inaccurate prediction of disparity map boundaries in stereo matching. Specifically, the existing deep - learning models have the problem of multimodal distribution when dealing with boundary regions, which leads to over - smoothing and thus affects the overall accuracy and consistency of the disparity map.

### Problem Description

1. **Multimodal Distribution Problem in Boundary Regions**
- In stereo matching, especially in boundary regions, determining which object a pixel belongs to becomes complicated, which often leads to multimodal distribution in the aggregated probability volume.
- This multimodal distribution will cause the over - smoothing problem, making the predicted disparity map inaccurate at the boundaries.

2. **Limitations of Existing Methods**
- Current methods mainly focus on four key steps: feature extraction, cost - volume construction, cost aggregation, and disparity regression, but the cost - aggregation module is not effective in dealing with local ambiguity.
- Especially in boundary regions, existing models have difficulty in effectively determining the归属 of pixels, resulting in inaccurate disparity estimation.

### Solution

To solve the above problems, the author introduced the concept of super - pixel soft constraints, aiming to improve stereo matching in the following ways:

1. **Utilizing the Consistency of Super - Pixels**
- The author observed that adjacent pixels are more likely to belong to the same object and show similar intensities within the probability volume of super - pixels.
- By introducing super - pixel segmentation, the network can generate a consistent probability distribution within each super - pixel, thereby improving the overall accuracy and consistency of the predicted disparity map.

2. **Improved Cost Aggregation**
- By integrating super - pixel information into the cost - aggregation process, the network is encouraged to generate a more consistent probability distribution and reduce the multimodal distribution problem.
- Use the Laplace distribution to model the true value and apply the cross - entropy loss function to suppress the multimodal problem, ensuring that the probability volume converges within the same super - pixel.

3. **Experimental Verification**
- The author carried out experimental verification on several commonly used datasets, and the results show that the proposed method can significantly improve the accuracy and consistency of the disparity map, especially in boundary regions.

### Summary

The main goal of this paper is to solve the problem of inaccurate prediction of disparity map boundaries in stereo matching by introducing super - pixel soft constraints, thereby improving the overall disparity - estimation precision and consistency.