Model Quality Aware RANSAC: A Robust Camera Motion Estimator

Shu-Hao Yeh,Dezhen Song

DOI: https://doi.org/10.48550/arXiv.1911.11884

2019-11-27

Abstract:Robust estimation of camera motion under the presence of outlier noise is a fundamental problem in robotics and computer vision. Despite existing efforts that focus on detecting motion and scene degeneracies, the best existing approach that builds on Random Consensus Sampling (RANSAC) still has non-negligible failure rate. Since a single failure can lead to the failure of the entire visual simultaneous localization and mapping, it is important to further improve robust estimation algorithm. We propose a new robust camera motion estimator (RCME) by incorporating two main changes: model-sample consistence test at model instantiation step and inlier set quality test that verifies model-inlier consistence using differential entropy. We have implemented our RCME algorithm and tested it under many public datasets. The results have shown consistent reduction in failure rate when comparing to RANSAC-based Gold Standard approach. More specifically, the overall failure rate for indoor environments has reduced from 1.41% to 0.02%.

Robotics

What problem does this paper attempt to address?

This paper attempts to solve the problem of robust camera motion estimation in the presence of abnormal noise. Specifically, the paper points out that although existing methods (such as those based on Random Sample Consensus (RANSAC)) have made certain progress in detecting motion and scene degradation, the failure rate of these methods in indoor environments is still relatively high, approximately 1.41%. This high failure rate may lead to the failure of the entire Visual Simultaneous Localization and Mapping (vSLAM) system, because a single failure may cause the continuous monocular vSLAM system to crash. Therefore, the paper proposes a new Robust Camera Motion Estimator (RCME), aiming to further reduce the failure rate by introducing model - sample consistency tests and inlier set quality tests, thereby improving the robustness of the algorithm.

### Main improvement points:
1. **Model - sample consistency test**: In the model instantiation step, verify the consistency between the model and the samples. This helps to exclude wrong models established due to over - parameterized sampling.
2. **Inlier set quality test**: Use differential entropy to verify the consistency between the model and the inlier set. First, determine the inliers according to the model uncertainty, and then verify the quality of the inlier set through the differential entropy test.

### Experimental results:
The authors of the paper implemented the RCME algorithm and tested it on multiple public datasets. The experimental results show that, compared with the traditional RANSAC method, RCME significantly reduces the failure rate on almost all test data. In particular, in indoor environments, the overall failure rate is reduced from 1.41% to 0.02%.

### Conclusion:
The RCME algorithm proposed in the paper effectively improves the robustness of camera motion estimation by introducing model - sample consistency tests and inlier set quality tests, and significantly reduces the failure rate, especially in indoor environments. This improvement is of great significance for improving the stability and reliability of the vSLAM system.