本发明专利技术公开了一种基于ORB‑SLAM2的双目稠密建图方法,涉及机器人同步定位与地图创建领域,该方法主要由跟踪线程、局部地图线程、闭环检测线程和稠密建图线程组成。其中稠密建图线程包含以下步骤:1)场景深度范围估计,限制立体匹配计算量;2)立体匹配,估计每个像素点的逆深度值;3)帧内平滑、外点剔除,增加逆深度图的稠密度的同时保留深度边界的不连续性;4)逆深度融合,对上一步获得的逆深度进行帧间优化;5)帧内平滑、外点剔除。本方法只利用CPU,实现室内、室外环境的稠密地图创建。
A binocular 3D dense mapping method based on ORB-SLAM2
The invention discloses a binocular dense mapping method based on ORB_SLAM2, which relates to the field of robot synchronous positioning and map creation. The method is mainly composed of tracking thread, local map thread, closed loop detection thread and dense mapping thread. The dense mapping thread includes the following steps: 1) depth range estimation, which limits the computation of stereo matching; 2) stereo matching, which estimates the inverse depth of each pixel; 3) intra-frame smoothing, exterior point elimination, which increases the density of the inverse depth map while preserving the discontinuity of the depth boundary; 4) inverse depth fusion, which is the last step. The inverse depth is used for inter frame optimization; 5) intra frame smoothing and external point elimination. This method only uses CPU to realize dense map building in indoor and outdoor environment.
【技术实现步骤摘要】
一种基于ORB-SLAM2的双目三维稠密建图方法
本专利技术属于机器人同步定位与地图