Research Of The Key Technology On Dimensional Measurement Based On Binocular Stereo Vision

With the development of China's industrial production, many new technologies have been applied, and the measurement technology has a very important role in the field of industrial production. The efficiency and accuracy of traditional measurement methods are relatively low, which can not meet the requirements of modern industry. The measurement methods which have high accuracy either low efficency or high cost of measuring equipment. Measurement technology based on computer vision not only has high measurement precision and low cost, so it is very important to study the measurement technology based on computer vision.This paper set up a test environment for stereo vision with a stereo camera. The camera calibration and stereo matching are studied, experiments are done by using the proposed method and the results illustrate that this method can reach desired requirement. The contributions of this paper include the following three aspects:Firstly, we analyze the status of some current mainstream measurement method and summed up their advantages and disadvantages respectively. we propose a method for Dimensional Measurement based on a stereo camera.Secondly, an improved method for stereo camera calibration based on image metadata is presented in this paper. The method uses the meta information in the photos taken with modern digital camera. After the feature point are extracted and matched, the 3D coordinates of the match points will be recovered. Then, the Bundle Adjustment can be used to obtain the optimized value of camera intrinsic parameter and matrix after the iteration during the process of geometric re-projection errors about the 3D points. This method has no special requirement on the scene. And the camera parameters will be constantly optimized in measurement process.Thirdly, the solution for the actual requirement of measure is using a matching method based on area template. This paper propose an improved process for stereo matching based on photometric feature and Census feature. To avoid the effects of noises and distortion, the central pixels' gray values of region template are treated. We can determine the gray value through taking the median of the template gray values.Finally, the effectiveness was validated by the experiments. These experiments are designed measure form different angles and distance. The result of the experiments show that the proposed approach here is effective, and the precision of measure can meet practical application.