采用混合回环检测与闭环优化的TSDF地图创建  

TSDF map building based on hybrid loopback detection and closed-loop optimization

作　　者：贾松敏 李柏杨 张国梁 JIA Song-min;LI Bo-yang;ZHANG Guo-liang(Faculty of Information Technology,Beijing University of Technology,Beijing 100124,China)

机构地区：[1]北京工业大学信息学部,北京100124

出　　处：《光学精密工程》2018年第6期1497-1506,共10页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.61175087);北京工业大学2017智能制造领域大科研推进计划资助项目(No.040000546317552)

摘　　要：由于原始TSDF(Truncated Signed Distance Function,TSDF)模型仅考虑相邻时间上的关联,误差将不可避免的累积到下一时刻,无法构建全局一致的地图。为了实时精确的建立大场景稠密3D地图,对TSDF模型进行了改进。首先,构筑相机位姿模型和加权融合3D点截断信息的TSDF模型,用于准确表示创建物体的表面。其次,提出一种改进的回环检测方法,并将其与随机蕨类彩色图像编码化相结合,进而优化TSDF模型,即混合优化位姿模型。最后,使用g2o图优化库解算约束函数,建立数据集间的优化边。实验结果表明:混合优化位姿模型能识别曾到达区域,特别在较大场景下使用可以得到更加准确的相机轨迹和地图。采用TUM数据集中的fr1/xyz、fr1/room、fr1/desk对所提算法进行检验,结果表明该方法能够使相机轨迹的均方根误差分别下降0.59cm,3.14cm,0.94cm。在室内环境和公开数据集上的实验结果证明了所提算法的有效性和准确性。Because the original Truncated Signed Distance Function （TSDF） model only considers the correlation of adjacent time, the error will inevitably accumulate to the next time, and it is impossible to build a globally consistent map In order to establish the dense 3D map of large scenes in real time and accurately, the TSDF model was improved in this paper. First, a camera pose model and weighted fusion 3D point truncation information model were established to represent the surface of the created object. Secondly, an improved loop detection method was proposed, which was combined with the random fern color image coding, and then optimized the TSDF model, that is, the mixed optimization pose model. Finally, the g2o diagram was used to solve the constraint function and established the optimized edges between the datasets. The experimental results showed that the loopback detection method proposed in this paper could obtain more accurate camera trajectory and map in a larger scene. The proposed algorithm was verified by fr1/xyz, fr1/room and fr1/desk in the TUM dataset. The results show that the root mean square error of the camera trajectory decreased by 0.59 cm, 3.14 cm, 0.94 cm, respectively. The validity and accuracy of the proposed algorithm are verified in the laboratory environment and the open data set.

关 键 词：随机蕨类 TSDF 闭环优化 图优化 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]