基于点线视觉/惯性SLAM和目标检测的测距方法

Ranging method based on point-line vision/inertial SLAM and object detection

针对无人平台在未知环境中自身定位和对远距离目标测距精度不高的问题,利用单目相机和惯性器件组成视觉/惯性导航定位系统,结合目标检测提出“动态基线+三角测距”方法实现自身定位和对目标测距。首先,建立基于点线融合的视觉/惯性系统模型,提高系统自身定位精度,给出运动前后相对位姿变化;其次,利用目标检测算法对目标进行检测和识别,得到运动前后物体对于图像平面的视差;最后,通过三角测距实现对目标的高精度测距。公共数据集实验测试结果表明,引入线特征的视觉/惯性系统的平均定位均方根误差(RMSE)为0.15 m。无人平台搭载系统对目标进行测距实验表明,系统在100 m以内对目标测距的误差小于测距距离的2%。

Aiming at the problem that the unmanned platform is self-positioning in an unknown environment and the accuracy of long-distance target ranging is not high,a vision/inertial navigation and positioning system is composed of a monocular camera and an inertial device,and the"dynamic baseline+triangulation ranging"method is proposed in combination with target detection to realize self-positioning and target ranging.Firstly,a vision/inertial system model based on point-line fusion is established to improve the positioning accuracy of the system itself and give the relative pose changes before and after movement.Secondly,the target detection algorithm is used to detect and identify the target,and the parallax of the object to the image plane before and after motion is obtained.Finally,high-precision ranging of the target is realized by triangulation.The experimental test results of the common dataset show that the average localization root mean square error(RMSE)of the vision/inertial system with line features is 0.15 m.The unmanned platform mounted system to conduct ranging experiments on the target shows that the error of the system in measuring the target within 100 meters is less than 2%of the ranging distance.