Design And Implementation Of Vision Positioning System Based On Convolutional Neural Network

With the rapid development of computer technology,automation and informationization of industrial production are more and more valued by people.In the process of development,machine vision technology has gradually become a research hotspot because of its high accuracy and non-contact characteristics.Using binocular vision target recognition and location technology in actual production process can greatly improve production efficiency and quality.With the development of convolutional neural network deep learning technology,industrial robots combined with machine vision technology are gradually used to help people deal with more and more complex processing problems.In this paper,the combination of the latest convolutional neural network model and machine vision technology is studied,and the research results are applied to target workpiece recognition and location scene.In this paper,two identical cameras are used to sense the environment,and the convolution neural network is used to complete the learning of image information.Finally,the target workpiece is recognized and positioned by binocular vision positioning principle.The main contents of this paper include the following aspects:(1)Several common camera imaging models and Zhang Zhengyou's calibration methods are studied.This paper introduces the methods to solve the internal and external parameters of different camera imaging models,derives the calibration process of Zhang Zhengyou calibration method under different camera models and the transformation relationship between coordinate systems in the calibration process,and finally solves the internal and external parameters of the camera in the system through experiments.(2)The network model,training method and over-fitting of convolutional neural network are studied.The role of different network structures in model training and the methods of parameter transmission and Optimization in network model are analyzed.(3)The SSD target detection algorithm is studied,and a multi-scale fusion based SSD algorithm is proposed.The relevant network model is constructed on VOC and MSCOCO data sets.The effectiveness of the proposed algorithm is verified by relevant training and testing.(4)On the basis of target recognition and detection,the related principle of binocular vision three-dimensional positioning is studied in depth.Using the results of target recognition and detection,the three-dimensional information of the target workpiece is reconstructed,and the mapping and derivation from the two-dimensional information of the workpiece image to the three-dimensional coordinates are completed.Finally,using the above research results,a visual positioning system based on convolution neural network is designed and implemented,and the system is tested.The experimental results show that the recognition and positioning effect of the system is stable and has high practical value.