Error Analysis And Control Of Measurement System Based On Binocular Vision

In the field of computer vision, it is very important to reconstruct the three-dimensional message according the two photos of the scene. The technology which imitates human vision is called binocular vision. Binocular vision is widely used in robotics, automatic tracking of the objects, scene recovery of the accident and three-dimensional reconstruction.The brain of human obtains images according to the eyes of us, and the images are processed to reconstruct the message of the scene according to the theory of binocular vision. The binocular vision theory uses the two cameras to imitate human eyes, and the two cameras take the photos of the same scene, then the computer will compute the three-dimensional message of the scene. The basic skeleton theory of binocular vision is pinhole model. We can idealize theory of binocular vision as pinhole model so that we can understand and analyze theory of binocular vision. From pinhole model, we can find that we can’t get the three-dimensional message using single image. We will get the depth message (we can get three-dimensional message of the point by the depth message according to the theory of binocular vision) of a point using the two image of different angle of the point. The method reconstructing the depth message by the imaging points of the point is called stereo vision method, and the process of stereo vision is called three-dimensional reconstruction. Three-dimensional reconstruction includes several steps. Each step, for example, generating image, match of feature points, camera calibration, affects the result of three-dimensional reconstruction greatly. Although the theory of three-dimensional reconstruction is more and more mature, there are still errors in three-dimensional reconstruction for the error of points match or the error of calibration or the error of other steps in practical application. How to control the error of every step is very important in practical application. This article mainly discusses the3D measurement system based on binocular vision and analyzes the factor of the errors in the system and the relationship between errors and each variable. In3D measurement system based on binocular vision, errors will impact greatly the result of measurement. In general, there exists two main categories of error in measurement system:one of them emerges in the procedure of calibration, and the other emerges in the procedure of measurement. Assuming that the result of calibration is precise, we mainly discuss what factors affect the accuracy of the measurement system and how the factors affect the accuracy of the measurement system. Based on the offset between the ideal and the real projection point on the CCD plane, we analysis relationship of the focal length, convergent angle,baseline distance and depth of the point to camera with the Measurement Error, and some experiments will be done to prove the error model of our theory. The aim of this paper is to decrease error of the measurement system based on binocular vision by setting the camera with the error model.The result of the experiment shows that the error distribution corresponds to the error model and the error model can be used to control the error of the measurement system. We do not discuss the error of calibration, so, the key point for the future research work is how to reduce the error of the calibration.