基于物体检测的视觉惯导SLAM联合优化
首发时间:2021-04-22
摘要:同时定位与地图构建(Simultaneous Localization And Mapping,SLAM)技术可以解决自我定位和环境感知的问题,其被广泛地应用在无人驾驶、机器人导航等领域中。目前的视觉SLAM技术大多是在像素或特征点级别进行定位与地图的构建。但在现实的环境场景中,除了这些几何特征还存在大量的物体信息,将这些物体信息融入到视觉SLAM技术中有利于增强视觉SLAM系统对环境场景空间结构的理解,是视觉SLAM技术的重要发展方向。本文基于物体检测过程,设计实现了物体级视觉惯导联合优化算法。通过惯导数据、视觉数据以及物体信息的误差构造了一个非线性优化问题,通过对优化问题的求解得到较优的状态估计结果。一定程度上提高了系统的鲁棒性和精确度,进一步加深了对场景的理解,实现了带物体标签的点云地图构建和大规模的系统定位,并通过实验验证了本文方法的有效性。
For information in English, please click here
The joint optimization of visual-inertial SLAM based on object detection
Abstract:Simultaneous Localization And Mapping (SLAM) technology, which mainly solves the problems of self-positioning and environmental perception, has been widely used in many fields such as Automatic Driving, Robot Navigation and other important scenes. However, most of the current visual SLAM technologies perform positioning and map construction at the pixel or feature point level. In real environmental scenes, there is a large amount of object information in addition to these geometric features. Integrating these object information into the visual SLAM technology can help machine to understand the spatial structure of the environmental scenes, which is an important development direction of visual SLAM technology. This paper proposes and constructs a visual-inertial and object information joint optimization algorithm based on object detection. A nonlinear optimization problem is constructed by the errors of inertial data, visual data and object information. Through solving the optimization problem, state estimation has been improved. The joint optimization algorithm has improved the robustness and accuracy of the system, and further deepened the understanding of the scene. The construction of point cloud maps with object tags and large-scale system positioning are realized, and the feasibility and effectiveness of the joint optimization algorithm are verified by experiments.
Keywords: monocular slam object detection visual-inertial slam
基金:
引用
No.****
动态公开评议
共计0人参与
勘误表
基于物体检测的视觉惯导SLAM联合优化
评论
全部评论0/1000