Research On Key Technologies For Videometrics Based On The Fusion Of Camera And Inertial Measurement Unit

In vision-guided allocation and transport,the challenges of limited computing resources,sparse scene features and a large number of repeated textures seriously affect the real-time,accuracy and robustness of videometrics.Inertial measurement unit(IMU)can measure attitude angle at high frequency.Integrating it with camera can simplify the complexity of visual measurement problem,reduce the dependence on image features and improve the reliability.Therefore,the research on videometrics method based on the integration of camera and IMU can provide theoretical and practical guidance for the accurate and intelligent transformation of dispatching operation,which has very important theoretical significance and engineering practical value.This thesis studies the key technologies for videometrics based on the fusion of camera and inertial measurement unit,including camera self calibration,camera and IMU rotation alignment,and IMU-assisted camera absolute pose estimation methods and applications.On this basis,a complete and applicable camera and IMU integrated vision navigation scheme is designed to overcome the above challenges.It provides theoretical and practical support for real-time and accurate absolute pose measurement for allocation and transport.The main achievements of this thesis can be summarized as follows:(1)Aiming at the problem of camera self calibration,a camera self calibration method based on affine correspondence and known relative rotation angle is proposed.In this method,the affine correspondence provides two additional constraints for the self calibration problem compared with the traditional point correspondence.The known relative rotation angle can be derived from IMU measurement data.Therefore,in the framework of Random Sample Consensus,fewer matching point pairs are required to calibrate the camera parameters,which can reduce the number of iterations and improve the speed and robustness of the algorithm.Experiments show that the camera self calibration method has high robustness and accuracy for mismatching caused by repeated texture and a certain degree of noise.(2)Aiming at the problem of camera and IMU rotation alignment,an accurate and easy to operate method for camera-IMU system’s rotation relationship calculation is proposed.This method uses a single affine correspondence to calculate the rotation matrix,which is the minimal case solution of the problem.The known initial rotation angle between camera and IMU is used to approximate rotation matrix in the first-order form.The calibration model can be simplified to polynomial equations based on homography constraints,reducing the computational resources required to solve the equations.In addition,through the joint optimization of multiple stereo image pairs,more accurate calibration results can be obtained.The proposed method does not need any additional auxiliary equipment or any specific motion of the camera,and can effectively suppress feature mismatch.The experiment results on simulated data and two real data sets show that the accuracy and efficiency of this method are slightly improved compared with the existing methods.(3)Two minimal case solutions for absolute pose estimation are proposed and implemented,and the corresponding visual navigation schemes are designed.(i)The plane moving target visual navigation method based on two 2D-3D matching point pairs(2D-3D point pairs for short)fully considers the relative attitude between the camera and the moving target body.It is suitable for the case that the camera faces any direction.It is a linear solution method with unique solution,and its minimum required matching point logarithm is 2.It can also directly process any point pair whose number is greater than 2.Experiment results show that this method can greatly improve the accuracy of pose estimation of plane moving targets and reduce the calculation time.(ii)The general moving target visual navigation method based on two 2D-3D point pairs and vertical direction does not restrict the target to move in the plane,and the vertical direction can be obtained directly by using IMU physical measurement.The experiment results show that the relative measurement error of this method is less than one thousandth.