Research On Low Contrast 3D Geometric Parameter Measurement Based On Binocular Vision

The strategy of 'Made in China 2025' has put forward higher requirements for the production quality of products.The birth of computer vision theory has opened up an innovative way for the size detection of products.When detecting targets in industrial production,the detected light intensity or the property of the target object itself causes the phenomenon of global or local low contrast in the collected images,which leads to the failure to accurately detect the outline and feature points of the object.In view of this phenomenon,research on measurement method of low contrast three-dimensional geometric parameters based on binocular vision is carried out,mainly completing the following aspects:Firstly,the background and significance of binocular vision research are introduced.By analyzing the development of binocular vision at home and abroad as well as the treatment of the phenomenon of low contrast of the target object,the overall research route is formulated.In this paper,the basic concept of binocular vision calibration and the calculation method of depth information are introduced.The internal and external parameters of the camera are obtained by using the structure of converging binocular vision system and Zhang Zhengyou's planar calibration plate method based on Python language.Secondly,the relevant algorithms of image preprocessing are introduced.Aiming at the problem that the existing image filtering algorithm will blur the edge in the filtering process,the improved image gradient guided window adaptive bilateral filtering is used for image denoising.According to the characteristics of low contrast,Retinex image enhancement algorithm was used to obtain high contrast image plane by using improved HSV channel.Thirdly,the matching primitive,matching criterion and similarity measurement function are introduced.Aiming at the mismatching problem of matching primitive,an improved feature extraction algorithm combining edge detection and FAST feature extraction was proposed.The subpixel coordinates of feature points were extracted by using polar constraint criterion,and the BRIEF descriptor was used for feature matching.Finally,combined with the actual application,the three-dimensional measurement of the relay module is carried out by using the binocular vision detection platform.The experimental results show that the proposed 3D measurement method can effectively measure the low contrast phenomenon of the target object,and can obtain a high measurement accuracy.