Research On Target Recognition And Location Algorithm For Machine Vision

Using machine vision technology to realize the detection and recognition of workpiece in production line can effectively improve the flexibility and intelligence level of manufacturing.As an important research direction of machine vision,the adaptability and real-time performance of target recognition and localization are important indexes for machine vision system.The process of target recognition includes image feature extraction,feature matching,and similarity or distance calculation.In the industrial field,the basic requirement for target recognition algorithm is adapting to the linear deformation of the target.Therefore,a target recognition algorithm with translation,rotation and scale invariance is required.Industry also has a high demand for real-time.In addition,the image may be affected by the noise due to the harsh industrial environment.The stability and robustness are also important ind icators.This paper proposes a target recognition algorithm at industrial applications that satisfies the translation,rotation and scale invariance,and has an adaptable ability to noise and nonlinear deformation.The main research contents are as follows:Firstly,based on the idea of outline shape segment,the recognition algorithm is divided into two parts: rough matching and fine matching.The similarity between shapes is weighted by the results of rough matching and fine matching,which has a better global and local description ability.Secondly,a feature point extraction method based on the minimum circumscribed rectangle is proposed to enhance the adaptive ability of the algorithm to noise and nonlinear deformation.The center of the smallest circumscribed rectangle represents the center of the target so that the constructed descriptors are more stable.Thirdly,in order to speed up the matching,by using the extended minimum edit distance matrix and the sequential information of the contour points,the matching relationship between the different contour segments is obtained.The fine matching only needs to be performed between matching contour segments,which reduces the number of matches and speeds up the overall algorithm.Fourthly,after the completion of the target recognition,the minimum circumscribed rectangle center is taken as the center of the shape.The rotation angle is calculated by the highest similarity degree matching contour.The center and the rotation angle serve as the target locat ion information.Experiments show that the adaptability and real-time performance of the algorithm can meet the requirements of general industrial application.It can be applied to flight marking system.