Research On UAV Hovering Pose Detection System Based On Binocular Vision

With the continuous improvement of the stability and control ability of UAVs,the accuracy of the equipped vision measurement or other measurement systems is expected to be widely used,such as direct measurement methods or measurement control points in industrial measurement,surveying and mapping and other fields.To ensure the accuracy of the data collected by drones in precision measurement and fault monitoring,a set of evaluation standards need to be developed to detect the stability of the drone when hovering,and break through its application limit in the field of precision measurement.Aiming at the evaluation of the hovering accuracy of drones,a light and concise airborne binocular vision drone hovering pose detection system is proposed,and a three-dimensional cooperative target is designed and manufactured.Two extraction and matching reconstruction algorithms based on sparse feature points are respectively proposed,and the SAC-IA+ICP point cloud registration method is optimized.Compared with the optical flow method,it realizes the hovering pose measurement of the UAV with faster speed,higher accuracy and stronger reliability.The main content of the article is as follows:1、Briefly introduces the principle of binocular vision measurement method,and focuses on the description matching method of sparse feature points.A feature point extraction algorithm based on known feature constraints is used,and a feature point fast matching reconstruction algorithm based on projection shape invariant constraints is written,with a reconstruction accuracy of 0.28 mm,which achieves fast and highprecision sparse feature points Three-dimensional reconstruction.2、Aiming at the high resolution and high precision requirements of UAV sixaxis attitude change measurement,an improved SAI-IA+ICP point cloud registration algorithm is proposed to calculate the six-axis attitude change.The traditional point cloud registration method has the problem of incompatibility between registration accuracy and speed.The method of artificially optimizing the number and location of feature points and considering the occlusion problem of the acquisition system is used to design and produce a hemispherical three-dimensional target.It is verified by simulation that the displacement and attitude angle of the target have registration accuracy of 1.1078 mm and 0.0463°,respectively.3、Aiming at the volume and weight restriction requirements of the UAV’s load system,this paper designs and builds a set of drone hovering pose detection system based on binocular vision,and calls Python＿Open CV programming development based on Linux,which realizes the collection system on the PC side.And this multi-mode image acquisition remote control,wireless communication distance can reach about30 m in an open environment.4、Using the above binocular posture measurement system,using the laser tracker and T-Mac measurement value as the true value,the six-axis posture change was experimentally analyzed.Compared with the optical flow method,the measurement time of this method is reduced by 60%,the Euclidean translation distance measurement error is reduced by 83%,and the attitude angle deflection measurement error is reduced by 87%.The measurement system is mounted on an unmanned aerial vehicle to analyze its hovering drift trajectory,and the experiment proves that the measurement data of the measurement system is true and effective,has high reliability,and can be used for error compensation of the detection system.