Font Size: a A A

Research On The Indoor Localization System Of Multi-Vehicle Motion By In-Look Multi-Camera

Posted on:2020-07-11Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y H LiFull Text:PDF
GTID:1362330614950804Subject:Control Science and Engineering
Abstract/Summary:PDF Full Text Request
As an important means to assist the development of complex flight systems,indoor localization is widely used in the development of multi-vehicle vehicles.The vehicle kinematic parameters obtained in the indoor localization process are very important test data,which can be used for the development of multi-vehicle collaborative control and navigation algorithm,the assessment of vehicle guidance control system and the evaluation of vehicle control system performance.Based on the development of multi-vehicle motion indoor localization vision system,a method of precise solution of vehicle motion parameters and coordinate system unified is proposed for the motion localization of multi-vehicle under the condition of in-look multi-camera layout.And we also discusses and studies the high-precision positioning of target,in-look multi-camera calibration and multi-point target tracking and recognition of spherical targets involved in the development of multi-vehicle motion indoor localization system.Firstly,the image of the spherical target perspective projection is analyzed for the high-precision positioning of the spherical center imaging point in visual positioning,and the projection model is established on the basis of which.In view of the sub-pixel positioning problem of the image edge of the spherical target,the blur effect of the imaging edge is considered,the sub-pixel positioning edge model is established,and the sub-pixel positioning of the sphere projection image edge is realized by the Zernike moment edge positioning algorithm.In view of the problem of sphere center imaging point localization error compensation,according to the spherical-center perspective projection imaging model,combined with the result of spherical imaging edge pixel positioning,the closed solution of sphere center imaging point positioning error compensation is derived to achieve high-precision positioning of the spherical cooperative target center imaging point.Secondly,a in-look multi-camera calibration method based on one-dimensional(1D)calibration target is studied in view of the calibration and three-dimensional(3D)reconstruction of in-look multi-cameras in multi-vehicle motion indoor localization system.Taking into account the characteristics of the in-look multi-camera motion localization system,the multi-camera calibration model is established,the hierarchical reconstruction strategy is adopted,the Euclidean length of the 1D calibration object is calculated and all the camera matrix in the system is recovered.Based on the uncertainty analysis results of the characteristic point imaging in different perspectives under multi-view conditions,the weight of the measurement information of different perspectives in the 3D reconstruction is determined and the accuracy of 3D reconstruction is improved.At the same time,using the 3D reconstruction results,the initial calibration results are refined and solved by iterative method,which further improves the accuracy of the camera calibration.Thirdly,a multi-target online tracking method that considers the consistency of the target displacement vector is studied in view of the 3D tracking and identification of cooperative targets in the field of multi-vehicle motion indoor localization.Based on the mathematical description and analysis of multi-target tracking problems,the data association problem is transformed into a two-dimensional(2D)linear distribution problem,and by analyzing the spatial and temporal consistency of the position and velocity of the vehicle,the data association cost matrix is solved,and the tracking of multi-vehicle trajectory is realized by Bayes filter method.For the target identification problem,under the assumption that the initial position of the vehicle does not coincide,the rank o f the Hankel matrix composed of the 3D trajectory measurement data of the feature point is solved,the range of similar trajectory judgment is reduced,the recognition efficiency is improved,the track after rough recognition is processed by Hausdorff dist ance,and the trajectory number of each vehicle is registered.Finally,in view of the motion estimation problem in the process of localization in the vehicle's motion,a multi-vehicle motion indoor localization vision system is established,and the characteristics of cooperative target point laying on the vehicle are considered,and a method of localization solution based on the combination of the global calibration and vector of the coordinate system is proposed.The initial value of the vehicle motion localization that has been associated with the cooperative target data is solved.At the same time,considering that the tracking trajectory of multi-target tracking algorithm is not unique,the cumulative error of the initial value of motion estimation obtained by the position assignment solution,in the framework of graph optimization theory,the graph model is established,and the motion estimation results of the vehicle are optimized globally nonlinearly in the whole measurement process,solve the motion estimation of global optimization values.
Keywords/Search Tags:vision measurement, indoor localization, edge blur effect, in-look multi-camera calibration, Displacement vector consistency, graph model
PDF Full Text Request
Related items