Sensitivity analysis of the error factors in the binocular vision measurement system

Abstract

A method based on uniform experimental design and the backpropagation (BP) neural network to analyze the sensitivity of measurement errors in the binocular vision system is proposed. Six main error factors that affect the measurement accuracy of the system are summarized according to the binocular vision principle. The measurement experiment was carried out based on uniform design. The mapping relationship between the total measurement error and each error factor was established by BP neural network. Then the sensitivity coefficient of each error factor can be calculated by this mapping. These sensitivity coefficients provide a measure of the degree of influence each factor had on the total measurement error. The results show that the sensitivity coefficients of lens distortion and the image noise are 0.287 and 0.243, respectively, which are significantly higher than the other error factors. This means that these two error factors need to be emphatically suppressed during the measurement process.