Research On Multi-sensor High-precision Calibration Method Based On Statistical Error

As a research branch in computer vision,multi-sensor calibration technology has important application value in many fields that require three-dimensional reconstruction.In computer vision,in order to obtain the three-dimensional geometric data of the object to be tested from the acquired image,it is necessary to establish a three-dimensional positional relationship between the image point and the surface of the object,that is,the visual sensor needs to be calibrated to obtain the imaging system model parameters,so the calibration accuracy is subsequent 3D data processing and reconstruction accuracy have a huge impact.In addition,when carrying out three-dimensional information acquisition of the full surface of the target object,it is necessary to build a multi-sensor vision system,and how the multi-sensor calibration process reduces the calibration error and improves the overall reconstruction accuracy of the system,making multi-sensor calibration extremely important.Based on the above content,this paper takes the camera as the research basis,starts from the research theory of binocular to multi-eye sensor calibration,makes statistical analysis of some errors such as camera error in the camera calibration process and camera distortion,constructe optimization function,to reducing the influence of errors in the calibration process,so as to improve the precision calibration of multi-sensors calibration.The main research content of this paper:(1)An algorithm for precise location of targets based on circular marker points is proposed.Aiming at the problem of positioning the circle projection deviation in the perspective projection transformation of the polka-dot calibration pattern,the causes of the center deviation are analyzed firstly,then the ellipse is fitted by the least squares method to determine the position of the center of the circle.Finally,the center deviation in the projection transformation is corrected by the ring compensation..The experimental results verify the feasibility and accuracy of the proposed method by comparing the direct detection of the dot markers and the calibration of the real target points of the calibration images.(2)A statistical error calibration algorithm for multi-sensor systems is proposed.Statistical analysis is carried out on various errors affecting the calibration accuracy of the system during the calibration process,and an optimization model is established to reduce the system calibration error.In the camera calibration,the error is first classified,mainly divided into pixel plane error and spatial error.Then the error objective function expression is constructed,and the camera parameters are optimized by the objective function to obtain high-precision calibration results.Finally,the calibration experiment of multi-visual sensor is carried out,and compared with other calibration methods in accuracy.The results show that the proposed method has better accuracy and robustness.(3)A method for longitudinal displacement measurement of rail based on multi·sensor three-dimensional reconstruction is proposed.Firstly,through the precise positioning of the identification points,the correction of partial errors in the calibration process is calculated,and relatively accurate multi-sensor calibration parameters are calculated.Then,based on the image depth information,combined with the calibration parameter results,the acquired rail image is reconstructed in three dimensions and the longitudinal displacement measurement experiment of the rail is performed.Good results are obtained by comparing the measured results with the actual displacement distance.