【摘要】 针对传统三维重建算法存在的漂移问题,提出了一种端到端的在线大规模三维场景重建算法。首先,使用一种在线估计策略来鲁棒地确定相机的旋转姿态,同时构建层次优化框架用于融合深度数据的输入。然后,依据相机的全局估计姿态对每一帧的信息进行优化,解除了算法对目标跟踪时间的限制,完成了对帧间关系对象的实时跟踪。试验结果表明:本文算法的平均重建时间为399ms,平均估计迭代最低点(ICP)次数为20,完成每帧变换的时间为100ms;系统对大规模场景的重建具有鲁棒性,且实时性较好,是一种具有对应关系稀疏特性、结构信息稠密特性和相机光照一致特性的实时三维重建算法。更多还原

【Abstract】 An end-to-end online large-scale 3 Dscene reconstruction method is proposed.This method uses robustness to estimate the rotation attitude of the camera and constructs a hierarchical optimization framework for the fusion of depth data input.Then,the information of each frame is optimized according to the global pose of the camera,and the algorithm limits the target tracking time and completes real-time tracking of the frame.Experimental results show that the average time to reconstruct the algorithm reaches 399 ms and the average number of estimated Iterative Closest Point(ICP)times is 20,which needs 100 ms to complete each frame transformation.The system is robust to the reconstruction of large-scale scenes and has better real-time performance.This method is a realtime three-dimensional reconstruction system with corresponding sparseness, dense structure information and camera illustration uniformity.更多还原