At present structured light systems are widely used in the fields of robot vision industrial measurement and 3D face recognition This research mainly uses a structured light stereo vision system to measure objects and generate a point cloud with three-dimensional information on X Y and Z axes and then use the point cloud to do 3D object inspection Before generating the point cloud the calibration parameters of two cameras must be created Two cameras need to find their own intrinsic parameters first Then we can find out the extrinsic parameters between the two cameras through stereo vision calibration The corresponding relationship on the image plane of the two cameras can be found through the phase-shifting patterns created by structured light and then we can use the calibration parameters and triangulation method to get the three-dimensional information of the surface of the object in the three-dimensional world coordinate Since the algorithm of the structured light system is very time-consuming this research also aims at the acceleration of the whole system and also improves on the unstable point cloud results Finally the deep learning model was imported into our system to replace our one-to-one search for disparity value Experimental results prove that the proposed method can achieve XY-axis accuracy of 0 028mm Z-axis accuracy of 0 01mm and the overall execution time accelerated from 33s to 7s In the end deep learning is used to achieve Z-axis accuracy of 1mm
3D Object Inspection System Using Deep Learning-Based Structured Light and System Stability and Speedup
家瑋, 葉. (Author). 2020
Student thesis: Doctoral Thesis