Binocular-vision-based On-line Multi-measurement Of Workpieces

The implementation of China manufacturing 2025 promotes the demand of high automation and even intelligence in industries.This thesis proposes to use the binocular vision to achieve automatically on-line multi-measurement of workpieces.The main issues regarding binocular vision for on-line measurement are as follows:On analysis on the key issues regarding the calibration accuracy of the binocular camera,a plane calibration board based on solid center point sign and matching is presented on the basis of classic Zhang Zhengyou's calibration method.The proposed method is analyzed and discussed in theory.Experimental results demonstrate that the proposed calibration board is superior to the traditional one.Meanwhile,an improved calibration method based on two-step RANSAC algorithm is proposed.The calibration procedure of the binocular camera is developed to achieve better calibration performance.To deal with the issue that Canny detector cannot achieve exact edge location while keeping good noise suppression capability,an improved Canny operator is presented by using the idea of edge-focusing.Considering the high computation of the Canny operator,the proposed Canny method is implemented in Gaussian pyramid to detect edge features of workpieces accurately and quickly.The performance of the proposed method is verified by experiments.On survey of common-used matching methods,a gray-level correlation edge matching method based on polar line threshold constraint is developed to match contour features of workpieces.Experimental results show that the proposed method has good performance in terms of speed,accuracy and robustness.An on-line multi-measurement system using binocular vision has been set up by integrating the aforementioned techniques.The target of the system is to deploy in industrial assembly lines for on-line measurement.To this end,a light source system has been designed and a synchronized external trigger circuit of the binocular cameras has been developed.Experimental results demonstrate that the proposed system has good performance in terms of measurement accuracy and repeated error.