论文部分内容阅读
为了提升复杂环境中双目视觉里程计的精度,提出一种考虑多位姿估计约束的双目视觉里程计方法。首先,分别建立匹配深度已知点与深度未知点的数学模型,将深度未知点引入2D-2D位姿估计模型,从而充分利用图像信息。其次,基于关键帧地图点改进3D-2D位姿估计模型,且结合当前帧地图点更新关键帧地图点,从而增加匹配点对数,提高位姿估计精度。最后,根据改进的2D-2D及3D-2D位姿估计模型,建立多位姿估计约束位姿估计模型,结合局部光束平差法对位姿估计进行局部优化,达到定位精度高且累积误差小的效果。数据集实验和实际场景在线实验表明,本文方法满足实时定位要求,且有效地提高了自主定位精度。
In order to improve the accuracy of binocular vision odometer in complex environment, a binocular vision odometer method considering multi-pose estimation constraints is proposed. First of all, a mathematic model matching depth unknown point and depth unknown point is established, and the unknown point of depth is introduced into the 2D-2D pose estimation model to make full use of the image information. Secondly, the 3D-2D pose estimation model is improved based on the keyframe map points, and the keypoint map points are updated with the map points of the current frame to increase the number of matching points and improve the pose estimation accuracy. Finally, based on the improved 2D-2D and 3D-2D pose estimation model, a multi-pose estimation constrained pose estimation model is established, and the local beam adjustment method is used to optimize the pose estimation to achieve high positioning accuracy and small cumulative error Effect. Data set experiments and actual scene experiments show that the proposed method satisfies the requirements of real-time positioning and effectively improves the accuracy of autonomous positioning.