Research On Wheel Matching And Recognition Method Based On Machine Vision

The car wheel is an important part of the car.It plays the role of supporting the car and bearing external forces during the driving of the vehicle.Therefore,the quality of the wheel will directly affect the safety of the vehicle.Before the wheels leave the factory,the wheels need to be tested for quality.Because before the quality inspection,the wheels need to be classified and operated by different wheel model,so it is necessary to obtain the wheel model.In view of the many shortcomings of the traditional way of manually identifying the model of the wheel,this paper explores a method for identifying the model of the wheel based on machine vision,implements an automatic wheel model recognition system,and completes the classification and recognition of the wheel model.In the identification of the wheel model,the difficulties are mainly the separation of the wheel image from the background image,the extraction of the wheel image features,and the feature matching.In view of the above difficulties,the following research work has been done:1)In the image preprocessing stage,this paper designs an algorithm flow for extracting the circular area of the wheel hub and smoothing the burrs.In order to solve the problem that the accuracy of the circle positioning using the hough circle transformation algorithm is insufficient,a wheel circular area extraction algorithm that adds a candidate circle screening process is proposed.First,the wheel image is detected by the canny algorithm,then the candidate circle is extracted using the hough circle transform algorithm,and the candidate circles are filtered to accurately extract the best outer contour circle.Finally,the wheel area is extracted using the circle.Experiments show that this algorithm can accurately separate the wheel and background.2)In the feature extraction stage,this paper designs an algorithm flow that comprehensively uses multiple wheel features.First,the wheel radius is extracted using the hough circle transform algorithm as the first matching feature,and then the feature point extraction algorithm is used to extract the local feature points of the wheel as the second matching feature.The matching speed and accuracy of SIFT algorithm,SURF algorithm and ORB algorithm were comprehensively compared through comparative experiments.Finally,the ORB algorithm was used to extract the local feature points of the wheel.3)In the feature matching stage,first use the wheel radius for rough matching,narrow the matching range,and then use the feature points for exact matching.In the process of feature point matching,for the case of mismatched points,this paper proposes to use RANSAC algorithm to purify the matched points to eliminate mismatched points.Aiming at the situation that the wheel model cannot be identified due to the failure of the feature point matching,a rotation angle adjustment algorithm process is proposed.Firstly,the best and second best rotation angles are found by rotating the hub image multiple times,then using the wheel image at these two rotation angles to perform feature point matching,and finally taking the maximum number of matching points in the two matching results for identification.4)Build a complete wheel identification system,including image acquisition and display functions,database management functions,and matching identification functions.Comprehensive identification experiments were carried out on the production line,and the experimental data verified that the identification system proposed in this paper has the advantages of high recognition accuracy and strong real-time recognition,which can meet the needs of industrial production.