An improved measurement model of binocular vision using geometrical approximation

Abstract

In order to improve the precision of a binocular vision measurement system, an effective binocular vision measurement method, named geometrical approximation, is proposed. This method can optimize the measurement results by geometrical approximation operation based on the principles of optimization theory and spatial geometry. To evaluate the properties of the proposed method, both simulative and practical experiments are carried out. The influence of image noise and focal length error on measurement results is discussed. The results show that measurement performance of the proposed method is manifested well. Besides, the proposed method is also compared with Bundle adjustment and least squares method in a practical experiment. The experiment results indicate that the average error, calculated by using the proposed method, is 0.076 mm less than Bundle adjustment's 0.085 mm, and only half of the least squares method's 0.146 mm. At the meantime, the proposed method enjoys a high level of computational efficiency when compared to Bundle adjustment. Since no nonlinear iteration optimization is involved, this method can be applied readily to real time on-line measurements.