基于新型多传感器融合策略的移动端双目视觉惯性SLAM闭环算法研究

Mobile visual-inertial SLAM loopclosure algorithm based on novel multi-sensor fusion strategy

研究了复杂环境下基于视觉同时定位与地图构建(SLAM)算法的移动端实时定位问题。该问题有如下几个难点:首先是移动端设备的计算资源受限,这对算法的优化与解算效率提出了更严格的要求;其次是测试场景的复杂多变,使得算法在低纹理及快速运动等情况下容易丢失目标;最后是实际应用时对系统的可拓展性要求较高,需要具备复杂场景下的适应性。针对上述问题,本文提出了面向移动端的双目视觉惯性SLAM算法,采用新型多传感器融合策略,通过将双目视觉图像和惯性测量数据进行紧耦合优化,设计了移动端回环检测算法,显著提升了系统的鲁棒性和可靠性。通过实验验证了所提方法的有效性,其在定位精度上超过了当前同类方法的最好结果,并开发了移动端的增强现实(AR)应用,以展示系统在真实场景中的效果。

This paper deals with the real-time localization of mobile visual simultaneous localization and mapping(SLAM) in cluttered environments.There are several difficulties in this task.Firstly,the limited computational resources lead to the requirements of the optimization and efficiency on the algorithm.Secondly,there are the cluttered and dynamic scenario.How to avoid the drift in low texture area and fast motion is the main difficulty.Finally,it requires good scalability,which can be accurately landed and has applicability in certain application domains.To tackle the above challenges,this paper proposes a stereo visual-inertial SLAM algorithm for mobile devices.A new multi-sensor fusion strategy that optimizes stereo visual terms and inertial measurement error in a tightly-coupled way is introduced.And the loop detection algorithm on the mobile devices significantly improves the robustness and reliability of the system.The effectiveness of the proposed method is evaluated through the intensive experiments,and the localization accuracy outperforms the state-of-the-art methods.Moreover,an augmented reality(AR) application is developed as an application of the system in the real scene.