基于平面约束的SLAM位姿优化算法
首发时间:2022-01-12
摘要:平面通常存在于人造场景中,可用于鲁棒定位。在这篇论文中,提出一个新的单目视觉惯性里程计系统,利用多平面的先验知识。提出了一种利用平面信息进行快速定位的视觉惯性平面PnP算法,并且结合语义分割网络实现对特定类型平面的识别和提取。采用基于重投影一致性的方法展开平面,对深度估计误差具有鲁棒性。结合改进的边缘化和滑动窗口策略,计算量大大降低。该VIO系统在KTTI数据集上和实际场景中都进行了测试,充分说明了该方法对于SLAM位姿优化的有效性。该系统可以达到非常有竞争力的准确性,并且在长的和有挑战性的序列上工作得很好。
For information in English, please click here
A Visual-Inertial Odometry with Plane Constraints
Abstract:Planes usually exist in artificial scenes and can be used for robust positioning. In this paper, a new monocular vision inertial odometer system is proposed, which uses the prior knowledge of multi plane. A visual inertial plane PnP algorithm for fast positioning using plane information is proposed, and combined with semantic segmentation network to recognize and extract specific types of planes. The method based on the consistency of re projection is used to expand the plane, which is robust to the depth estimation error. Combined with the improved edge and sliding window strategy, the amount of calculation is greatly reduced. The VIO system is tested on KITTI data set and actual scene, which fully shows the effectiveness of thisA Visual-Inertial Odometry with Plane Constraints method for slam pose optimization. The system can achieve very competitive accuracy and work well on long and challenging sequences.
Keywords: Visual inertial odometry bundle adjustment Plane priors Reprojection consensus
基金:
引用
No.****
同行评议
勘误表
基于平面约束的SLAM位姿优化算法
评论
全部评论