摘要
针对Kinect Fusion算法中存在的重建范围小、缺少有效的重新定位策略及累计误差问题,提出了一种基于随机蕨编码的三维重建方法。应用随机蕨编码构建相机路径回环的检测策略减少长时间重建所产生的累积误差,通过检索相似关键帧进行相机位姿估计失败后的重新定位,通过与程序集成框架InfiniTAM相结合,增大重建范围。采用RGB-D SLAM验证数据集进行了对比实验。实验表明:提出的方法可以大大增加重建范围,在相机定位失败后有效地进行重新定位,同时减少了长时间重建产生的累积误差,使得三维重建的过程更加稳定,获得的相机位姿更加精确。
Aiming at problem of small reconstruction range,lack of effective relocalization strategy and cumulative error in Kinect Fusion algorithm,propose a 3 D reconstruction method based on random fern coding. The camera trajectory loop closures is detected by using the random fern encoding to reduce the accumulated error caused by long time reconstruction. By retrieval similarity key frames,relocate camera pose estimation after failure. By combining with Infini TAM,range of reconstruction are increased. A comparison experiment is carried out using the RGB-D SLAM dataset. Experimental results show that the proposed method can greatly increase reconstruction range,effectively relocate the camera pose after failure,and reduce accumulated error,which makes the 3 D reconstruction process more stable and obtains camera pose more accurately.
出处
《传感器与微系统》
CSCD
2017年第12期157-160,共4页
Transducer and Microsystem Technologies
关键词
随机蕨编码
三维重建
相机路径回环
关键帧
random fern encoding
3D reconstruction
camera trajectory loop closures
key frames
