基于双目立体视觉SLAM的林下实时定位

[目的]在林下环境中受到树木冠层的遮挡,易造成全球导航卫星系统卫星信号失锁,不能满足林下实时定位的需求.本研究引入双目立体视觉SLAM技术,在移动过程中根据周围环境中路标点的位置变化实时确定传感器的空间位置和姿态,以期在林下环境中达到实时定位的目的.[方法]实验过程中使用双目相机按照10 Hz的采集频率对3块边长为40...

Real-time positioning in forests based on binocular stereo visual SLAM

【Objective】Covered by tree canopy in the forest environment,it is easy to cause the Global Navigation Satellite System(GNSS)satellite signals to lose lock,which cannot meet the needs of real-time positioning in forests.This study uses binocular stereo vision SLAM technology to determine the location and orientation of the sensor in real time according to the changes in the position of the landmarks in the surrounding environment during the movement,in order to achieve the purpose of real-time positioning in forests.【Method】During the experiment,a binocular camera was used to continuously shoot three square plots with a side length of 40 m at a collection frequency of 10 Hz,and a representative landmark point was selected from the image according to the ORB feature extraction method.Then,the spatial position of the landmark was restored according to the imaging model of the binocular camera and the camera’s pose(location and orientation)were estimated in real time from the change in position of the landmarks.Finally,the BA(Bundle Adjustment)method was used to optimize the camera pose and the coordinates of the landmark points to further improve the positioning accuracy of the algorithm.The root mean square error of the control point coordinates in the acquisition route was calculated by comparing the measurement results of the total station to verify the positioning accuracy of the binocular stereo visual SLAM algorithm in forests.【Result】The experiment quantitatively analyzed the measurement results of the binocular stereo visual SLAM algorithm from the control point coordinate positioning accuracy and algorithm real-time performance.The results showed that the camera trajectory calculated by the binocular stereo visual SLAM algorithm and the total station measuring trajectory were basically the same.The root mean square error(RMSE)of control point coordinates along X-axis,Y-axis,and Z-axis direction were 0.81,0.65,and 0.58 m,respectively.The processing time of each frame was concentrated between 0.06 s and 0.08,and the processing speed was higher than its acquisition frequency.【Conclusion】The binocular stereo visual SLAM algorithm proposed in this study determines the pose of the camera sensor in real time during the movement,solves the problem of real-time positioning in forests,and lays the foundation for the forest plot survey by mobile acquisition.