Research On Wheel Vision Positioning And Recognition Based On Feature Dimension Reduction And Fusion

Machine vision technology has developed rapidly in recent years,and is currently widely used in industrial production fields,such as non-contact measurement of wheels,online positioning recognition,and so on.The wheel's positioning and identification system is based on machine vision technology to complete the extraction of wheel information,the advantage is that the position of wheel in the image plane can be known in real time,and the model of the wheel can be identified at the same time.This paper combines image processing and machine vision technology to complete the positioning and identification of different types of wheels,the positioning and identification methods are optimized to improve the efficiency and accuracy of wheel positioning and identification.The research work has certain theoretical significance and engineering application value.The main work and conclusions of the paper are as follows:(1)Research on wheel positioning method: First of all,determine the wheel positioning to extract the contour circle and key points of the wheel,the huff gradient method is used to extract the outer contour circle of the wheel,and then use the simplified and reduced SIFT-feature point matching positioning to find the space transformation matrix of the detected wheel and the template wheel,which is aimed to calculate the coordinates of the wheel's key points.Experimental results show that this method effectively overcomes the impact of uneven illumination on positioning,and is also robust to translation rotation and partial loss.(2)Research on wheel recognition methods: Collect seven different types of wheel pictures,and then pre-process the wheel pictures,including image denoising,image segmentation,morphological processing and edge detection,the characteristic values of the wheel hub are extracted on the basis of preprocessing,such as wheel radius,hole characteristics,spoke characteristics and invariant characteristics,etc;Then,serial,parallel,and weighted fusion methods are used to fuse wheel eigenvalues,and weights are set so that different eigenvalues occupy different proportions in the weighted final value to improve wheel classification accuracy;Finally,template matching and nearest neighbor classification Method and set threshold method are used to identify the three fused hub feature values,and a small number of samples are taken to make the ROC curves of the three classification models;The recognition results show that the improved weighted fusion method has higher recognition accuracy than serial and parallel fusion,and the recognition accuracy is 99%.According to the ROC curves,the classification model with the threshold method is optimal.(3)Complete the construction of the wheel positioning and identification system: Firstly complete the selection of the hardware of the wheel positioning and identification system,including computers,industrial cameras and light sources.Then use HALCON machine vision software platform and Visual Studio 2017 to complete programming and development of the wheel positioning and identification system software,the system will be used to locate and identify the input wheel images.After verification,the system can meet the requirements of wheel positioning and identification.To sum up,this paper researches the method of wheel positioning and identification,and carries out an in-depth discussion on wheel positioning and model identification.It proposes a method of wheel positioning with high robustness and high accuracy,and wheel identification with high accuracy,and the wheel positioning and recognition system was established to complete the intelligent positioning and recognition of wheel images.