Forest Tree Positioning Based On Stereo Vision And Inertial Fusion

In the forest resource survey,forest tree positioning is an important task,which is related to the establishment of fixed plots,the determination of forest stand density,and the mapping of forest maps.In the current forest tree positioning method,although the positioning method based on the fixed measurement point is highly accurate,the trees that are far away from the measuring point cannot be located due to mutual occlusion between the trees;The GPS-based forest tree positioning method can locate trees at any place,but due to the interference of shade blocking in the forest area,high positioning accuracy cannot be obtained.Therefore,in the aspect of forest tree positioning,it is urgent to introduce new technologies to break through the current technical bottlenecks.In the field of mobile robots,the SLAM technology combining visual and inertial navigation can realize robot self-positioning and large-scale environmental map construction with high precision and robustness.The problem of forest tree positioning in forest resource survey can be understood as a map construction problem with forest coordinates as map elements.In order to realize large-scale and high-precision forest tree positioning,this paper draws on the SLAM technology of visual and inertial fusion in the field of mobile robots,and studies a forest tree positioning method based on stereo vision and inertial fusion.Firstly,in order to realize the SLAM system of stereo vision and inertial navigation fusion,the binocular stereo vision SLAM technology based on feature point method is studied.After focusing on the feature extraction and matching,Perspective-n-Point solution,Bundle Adjustment optimization and loopback detection,the open source project ORBSLAM2 was transplanted and tested,and the realization of the binocular stereo vision SLAM system was completed.Secondly,the fusion with the inertial navigation is performed on the basis of the binocular stereo vision SLAM system.After establishing the kinematics model of inertial navigation,the IMU pre-integration technique is introduced for the problem that the IMU sampling frequency and the camera sampling frequency are inconsistent,which leads to the difficulty of fusion of visual and inertial navigation.The fusion of binocular stereo vision SLAM and inertial navigation is realized by jointly optimizing the IMU pre-integral error and the visual re-projection error.The field comparison experiments show that the binocular stereo vision SLAM system after the fusion of inertial navigation under the same experimental conditions has a significant improvement in the accuracy and robustness of pose estimation under pure visual conditions.Finally,in order to realize the SLAM system combining binocular stereo vision and inertial navigation for forest tree positioning,this paper focuses on the idea of adaptive threshold tree trunk identification based on color features.With the binocular camera pose provided by the SLAM system in real time,the coordinates of the identified trunk in the camera coordinate system are transformed into coordinates in the reference coordinate system to achieve global positioning of the tree.A forest plot experiment of 30m x 30m was selected for the forest tree positioning experiment,the experimental results show that the proposed method can locate the trees in any part of the forest,there is no problem that the forest tree positioning method based on fixed measurement points cannot be located on any tree by the mutual occlusion of trees.In terms of positioning error,the maximum error is 2.08m and the minimum error is 0.15m.The average error is 0.96m,which is less than the GPS-based forest tree positioning method.The proposed method of forest tree positioning is a scheme that compromises the positioning range and positioning accuracy,and provides a new technical idea for the forest tree positioning problem in the field of forest resource investigation.