单目SLAM增强现实测树系统设计与试验  

Design and Experiment of Monocular SLAM Augmented Reality Tree Measurement System

作　　者：范永祥 冯仲科[2] 闫飞[2] 申朝永[3] 关天蒴 苏珏颖 FAN Yongxiang;FENG Zhongke;YAN Fei;SHEN Chaoyong;GUAN Tianshuo;SU Jueying(Jihua Laboratory,Foshan 528000,China;Precision Forestry Key Laboratory of Beijing,Beijing Forestry University,Beijing 100083,China;The Third Surveying and Mapping Institute of Guizhou Province,Guiyang 550004,China)

机构地区：[1]季华实验室,佛山528000 [2]北京林业大学精准林业北京市重点实验室,北京100083 [3]贵州省第三测绘院,贵阳550004

出　　处：《农业机械学报》2023年第8期259-266,共8页Transactions of the Chinese Society for Agricultural Machinery

基　　金：广东省基础与应用基础研究基金区域联合基金——青年基金项目(2020A1515110253)。

摘　　要：将内嵌有面阵相机及IMU的智能手机作为硬件系统,单目SLAM技术获取多视图几何深度图、位姿等为数据源,构建了单目SLAM增强现实森林测树系统。设计了基于平滑度高鲁棒性过滤胸高圆柱体表面点云及切线的方法;然后,基于点到圆柱体表面距离及圆柱体切线到圆柱体表面距离构建了胸径与立木位置精确估计算法;最后,以该算法为基础在智能手机端开发了增强现实测树系统,即利用智能手机实时测树、并通过增强现实场景实时人工监督测量结果。新型测树系统在5块32 m×32 m方形样地中进行了测试,以评估新型测树系统的测量精度;此外,每块样地使用了单次观测、正交观测、对称观测及环绕观测4种不同的观测方法对立木胸高圆柱体观测,以评估不用观测方式对测树精度的影响。结果显示:立木位置估计值在X、Y轴方向的平均误差范围为-0.014~0.020 m,X、Y轴方向均方根误差范围为0.04~0.08 m;立木胸径估计值偏差为-0.85~-0.03 cm(相对偏差为-3.60%~-0.04%),均方根误差为1.32~2.51 cm(相对均方根误差为6.41%~12.33%);相比于单次观测方法,其他观测方法获取位置及胸径估计精度均有提高(特别是不可近似为圆柱体的立木树干),从精度与效率角度而言,正交观测及对称观测为最佳观测方法。结果表明,单目SLAM增强现实测树系统是一种可精确进行森林样地调查的潜在解决方案。A monocular SLAM forest test system was constructed,which used an ordinary smart phone embedded with an area array camera and an IMU as the hardware system,and used monocular SLAM technology to obtain depth maps,poses,etc.as data source.A method was designed based on smoothness and high robustness to filter the point cloud and tangent on the surface of the chest height cylinder;then based on the distance from the point to the surface of the cylinder and the distance from the tangent of the cylinder to the surface of the cylinder,an accurate estimation algorithm for the diameter at breast height and the standing tree position was constructed.Finally,based on the algorithm,an augmented reality tree measurement system was developed on the smart phone side,that was,real-time tree measurement by smartphones,and real-time manual supervision of the measurement results through augmented reality scenes.A highly robust method for filtering the point cloud and tangent on the surface of the breast height cylinder through smoothness was designed;then an accurate estimation algorithm for DBH and standing tree position was constructed based on the distance from the point and tangent to the surface of the cylinder.Finally,an augmented reality tree measurement system was developed on the smart phone end,which used the smartphone to measure trees in real time,and real-time manual supervision of the measurement results through the augmented reality scene.The tree measurement system was tested in five 32 m×32 m square plots to evaluate the usability of the tree measurement system at the same time;more importantly,each sample plot was investigated through single observation,orthogonal observation,symmetrical observation and surrounding observations in order to evaluate the impact of different observation methods on the accuracy of tree measurement.The results show that the deviation of the estimated standing tree position in the X and Y axes directions was-0.014~0.020 m,and the root mean square error range was 0.04~0.08 m;the deviat

关 键 词：森林 样地调查 单目SLAM 智能手机 

分 类 号：S758.7[农业科学—森林经理学]