Research On The Accuracy Theory And Experiment Of Large-Scale Vision Measurement

Measurement is an important method for human beings to understand the world and is the basis of science development. With the development of manufacturing industries, the large-scale, high accuracy, even all attitude and dynamical measurement are required. The traditional three-dimensional measurement technology (CMM) can not meet these requirements. In this context, the vision measurement technology emerges and is becoming more important. Based on the rigorous theory and modern software and hardware facilities, vision measurement system has been closely combined with computer. The vision measurement technology as a non-contact, high accuracy, rapid measurement method has been applied widely. Accuracy is the core indicator of any measurement task, so it is significance to study the accuracy theory of vision measurement.Supported by relevant fund project, the accuracy theory of vision measurement is studied. In this paper, the research covered measurement error sources, error propagation analysis, accuracy evaluation and camera calibration. The dissertation conducts researches on the following issues:(1)In order to derive mathematical formulas from the uncertainty of reconstructing points, projection matrix and the uncertainty of image points, the vision measurement theory based on the parameter equations is proposed. In this paper, the original projective equations are substituted by parameter equations, and the solved fomulas are rederived. This method overcomes the original disadvantages. By this method the error propagation analysis is simplified and the mathematical formulas of uncertainty of reconstructing points are provided clearly. Based on the formulas, the weighted least square algorithm for vision measurement is proposed. Simulation and experiment results show that the method in the conditions of small number cameras is better than the original method, and in the conditions of large number cameras is not worse than the original method.(2) For the small sample size of vision measurement, error atypical distribution, and not all factors can be expressed by formulas, the grey system theory is used to vision measurement system analysis. Firstly, some part of the grey theory is researched and improved in this paper. The grey coefficient vector is proposed to describe the uncertainty of vision measurement that increases the stability. Then the grey system relation analysis method is used to study the impact of those factors. The grey theory is used to research the vision measurement system to make up some shortcoming of the statistical method and the correlation between the measurement accuracy and factors is described. This method provides a reference for the vision measurement plan design.(3) In order to evaluate the measurement accuracy of large scale vision measurement system, the method named "the variance of fitting point" based on distance constraints was proposed. Firstly, several line segments of known length with common vertex were placed in the view field and these length values were considered as true values. In vision measurement system coordinate, these line segments could not cross one point because of the error existence. According to the length constraint, a point would be fitted by the least error square algorithm. Finally, the measurement accuracy was denoted by the variance of the fitting point. In the experiment, bias estimation for 88 points has one point bias beyond the limit deviation, and the uncertainty evaluation is consistent.(4) The camera calibration is key factor of vision measurement. Focusing on the camera calibration of the large scale vision measurement system, the calibrating method based on two-dimensional connectible flexible object is proposed by this paper. The method based on many two-dimensional objects that include extending part and discriminative part. The discriminative part composed by encoded points. Many sub-objects connected by those encode points that make up of a large size two-dimensional object. This method can connect sub-object without errors, then calibrating the camera with high precision. The method can extend the two-dimensional object area enough large, and it is flexible and convenience. The method can accomplish the camera calibration task of the large scale vision measurement system.