基于RGB-D SLAM手机的森林样地调查系统研究  

Research on Forest Plot Survey System Based on RGB-D SLAM Mobile Phone

作　　者：范永祥[1] 冯仲科[1] 陈盼盼 高祥 申朝永 FAN Yongxiang;FENG Zhongke;CHEN Panpan;GAO Xiang;SHEN Chaoyong(Precision Forestry Key Laboratory of Beijing,Beijing Forestry University,Beijing 100083,China;College of Science,Anhui Agricultural University,Hefei 230036,China)

机构地区：[1]北京林业大学精准林业北京市重点实验室,北京100083 [2]安徽农业大学理学院,合肥230036

出　　处：《农业机械学报》2019年第8期226-234,共9页Transactions of the Chinese Society for Agricultural Machinery

基　　金：中央高校基本科研业务费专项资金项目(2015ZCQ-LX-01);国家自然科学基金项目(U1710123);安徽农业大学青年基金重点项目(2015ZD06)

摘　　要：基于RGB-D SLAM手机构建了森林样地调查系统,该系统实现了样地构建、每木检尺及林分/样地参数的估计功能,并在测量过程中使用增强现实展示测量结果,且提供了重新测量的交互方式,使观测者在观测过程中能够检测结果的可靠性,并保证所获取样地信息的完整性。该系统在18块半径为7.5 m的圆形样地中进行了测试。结果显示,平均胸径估计值的偏差(BIAS)及均方根误差(RMSE)分别为0.36、0.69 cm,平均树高估计值的BIAS及RMSE分别为0.06、0.63 m,蓄积量估计值的BIAS及RMSE分别为8.595 9、25.735 8 m^3/hm^2,横断面积估计值的BIAS及RMSE分别为0.949 7、1.987 3 m^2/hm^2,株树密度估计值的BIAS及RMSE分别为-3、13株/hm^2,坡度估计值的BIAS及RMSE分别为0.30°、0.88°,坡向估计值的BIAS及RMSE分别为-0.44°、7.61°。其中,坡向估计具有较大的RMSE,是由于当坡度较小时,即使SLAM系统估计位姿有较小漂移,仍会导致该值产生较大偏差,但整体而言坡向仍是无偏的。Forest resources have their own importance in human survival and development.Forest plot survey is used to obtain forest information and analyze the status of forest resources.With the advancement in sensor technology,remote sensing,especially LiDAR,is used to obtain point cloud data by scanning plots which can be used to extract forest-based factors.The improvement of SLAM algorithm enables the positioning without GPS signal coverage.So that,the combination of LiDAR and SLAM system can be used to get a globally consistent point cloud of a plot under the canopy which can ensure the integrity and accuracy of the extracted plot properties.However,the estimations can not be checked and the omissions or errors can not be corrected.A plot survey system based on RGB-D SLAM mobile phone was developed,which constructed the process of plot survey,the estimation of tree-based properties and forest-based properties.Augmented reality technology was used to show the observer estimation results and the way of re-estimation,which ensured the reliability and integrity of the acquired plot information through human intervention.The system was tested in 18 circular plots with radius of 7.5 m.The average DBH estimations showed 0.36 cm BIAS and 0.69 cm RMSE;the average tree height estimations showed 0.06 m BIAS and 0.63 m RMSE;the volume estimations showed 8.595 9 m^3/hm^2 BIAS and 25.735 8 m^3/hm^2 RMSE;the cross-sectional area estimations showed 0.949 7 m^2/hm^2 BIAS and 1.987 3 m^2/hm^2 RMSE;the stem density estimations showed-3 stems/hm^2 BIAS and 13 stems/hm^2 RMSE;the slope estimations showed 0.30°BIAS and 0.88°RMSE;and the aspect estimations showed-0.44°BIAS and 7.61°RMSE.The aspect estimations had a large RMSE due to the estimated pose errors of the SLAM system,but the aspect measurements were still unbiased as a whole.

关 键 词：森林 样地调查 ToF相机 SLAM 增强现实 

分 类 号：S758.7[农业科学—森林经理学]