Research On Multi-vision Three- Dimensional Measurement

Machine vision technique has been extensively used in many fields,such as aviation,aerospace,war and civilian use.Especially,the multi-vision technology used for online three-dimensional(3D)coordinate measurement tasks for reflective markers on the target object can be seen everywhere,such as,motion capture and 3D geometry measurement of large scale object,et al.Owing to the advantages of visual measurement system,such as simple structure,convenient on-site installtion,higher stability and expansibility,in recent years,its related applications develop rapidly.Meanwhile,higher accurcy of 3D coordinate measurement is demanded,which urges to match the eponymic iamge points in multi-view more accurately and to establish better high-precision multi-vision measurement model,camera calibration model,and relative position estimation of multi-camera model.In this paper,we present the research status of related technologies in multi-vision system and summarize the exsiting problems.Focusing on the problems,we did the works as following:Firstly,we built a high-precision multi-vision measurement model.We defined the measurement coordinate system and introduced a basic optimization algorithm into it.We proposed a generalized orthogonal projection method to compute linear closed-solution to solve the space forward intersection problem,which provides a good initial value for the weighted nonlinear optimization.To match the eponymic image points without imaging differences,we proposed a new geometric constraints based matching method,which contains foundamental matrix,trifocal tensor,and the shortest distance from the projection point to the rest of the projection rays acooding to the number of effective cameras.Then,we built a camera calibration model.First,the camera related basic theoretical knowledge was elaborated.Then,closed solution of camera intrinsic parameters is obtained by using trigonal elimination point method based on 3D virtual calibration template and radial distortion correction model is established.Multi-point optimization in fitting straight line method and objective function based on Mahalanobis distance was adopted.Because a large number of feature points are needed in the calibration optimization process,and straight lines need to be fitted,this process is susceptible to measurement errors.Finaly,we propose a multi-camera relative position estimation scheme.The absolute coordinate system is established by using four-point L-frame.The initial value of camera relative absolute coordinate system was obtained by inverse 1D calibration idea and geometric constraints.The optimal result of free motion of 1D wand in measurement space was obtained and the azimuth between two cameras is calculated directly.And,the position fixing constraint was introduced for consistency correction.Combining previous results full use of redundant information of multi-camera,parameters' global optimization achieved.Synthetic data verify the strong robustness and high precision of the proposed methods.The experimental results demanstrate that the proposed algorithms can meet the demands of multi-vision measurement.