| 摘要: |
| 卫星/惯性/视觉多源导航因其精度和鲁棒性的优点，逐渐成为自动驾驶等智慧交通领域中的主流导航方案。视觉信息是这一多源导航系统中的关键输入之一。然而，视觉信息存在误差模型随场景变化显著、故障率高等问题。为保障导航定位的可靠性，增强视觉导航信息的完好性，提出了基于运动约束的视觉前端故障检测器，并与随机采样一致性（Random Sample Consensus,RANSAC）结合来降低视觉信息故障率。同时，基于误差包络理论确定保守的视觉前端误差模型。实验结果表明，该视觉前端完好性增强方法能够明显降低视觉特征点的误差标准差，相比于仅使用RANSAC的方法降低约70%。更重要的是，结合运动约束与RANSAC方法能够使得误差模型对场景不敏感。此项研究对于增强视觉参与的多源导航完好性具有重要意义。 |
| 关键词: 自动驾驶 增强视觉导航 完好性 多源导航 |
| DOI：10.20079/j.issn.1001-893x.230427001 |
|
| 基金项目:浙江省重点研发计划（2021C01196） |
|
| Visual Frontend Integrity Enhancement and Error Modeling with Integrated Motion Constraints and RANSAC |
| ZHAO Liang,WAN Zixiang |
| (Zhejiang Geely Space Technology Co.,Ltd.,Shanghai 200233,China) |
| Abstract: |
| Satellite/inertial/visual integrated navigation has gradually been becoming the mainstream navigation solution in intelligent transportation applications.To ensure navigation safety,it is necessary to realize integrity monitoring of this type of multi-sensor integrated navigation system.To achieve this goal,an outlier rejection method is proposed for visual measurements based on the motion constraint,which is combined with the Random Sample Consensus(RANSAC) method to further reduce the outlier rate.Finally,The error model for the visual measurements is established by employing the error overbounding technique.Experimental results suggest that the proposed method can significantly reduce the error standard deviation of visual feature points,which is about 70% lower than using RANSAC method alone.More importantly,the method combining motion constraints with the RANSAC method can make the error model insensitive to the scene.The research is of great significance for enhancing the integrity of multi-source navigation through visual participation. |
| Key words: autonomous driving enhanced visual navigation integrity multi-source navigation |
