Research On The Method Of Recognition Tire Marking Points Based On Machine Vision

The dynamic balance and uniformity of tires are two important factors affecting the safety and comfort of automobiles.Color and shape difference of marking points printed on tires indicate the dynamic balance and uniformity of the tire,which play an important role in tire installation.It is an important part of quality inspection for finished tire to check whether marking points are correct.However,the traditional manual recognition is too intensive and also it is not efficient and accurate enough.In order to improve the accuracy and efficiency of tire marking point recognition,an online recognition system of tire marking points based on machine vision technology is designed and the recognition method of the marking point is also studied in this paper.It mainly includes:According to the demands of tire marking point online recognition,the overall solution for online recognition of marking points based on machine vision was proposed and the visual recognition module is designed and analyzed detailed.In the hardware section,the visual imaging scheme is designed according to the features of the detection scene,and key components are selected;In the software section,the corresponding image processing algorithms are designed according to the collected tire images.The corresponding communication scheme is provided for the online recognition system to ensure the accurate and timely data transmission.The features of tires in the original image acquired by industrial camera are analyzed,the methods of inner circle positioning based on minimum enclosing circle and gradient CHT are proposed.The experimental results show that the inner circle positioning method based on gradient CHT has higher positioning accuracy and better robustness.On the basis of positioning of the tire inner circle,according to the ratio of tire inner diameter to outer diameter in MES information,the tire outer circle can be determined.Furthermore,the accurate positioning of the tire is realized,which is the foundation of the subsequent marking points recognition.According to the color feature of marking points,the red and yellow candidate marking points are obtained by the back projection of H-S color histogram;Using the white marking point feature of high luminance,the white candidate marking points are obtained by the threshold segmentation of the grayscale image in the luminance channel.A deep convolutional neural network model is built based on residual learning structure to classify candidate tire marking points,which can recognition the marking point and its color;Taking the red marking points as an example,in order to solve the problem of over fitting problems caused by unbalances sample types and the difference between training samples and test samples,six adaptive normalization preprocessing methods are proposed to preprocess the marking points,the problem of over fitting of network model in shape recognition is solved,and the shape recognition accuracy is improved.The calculation method of marking point integrity is studied.the images are enlarged firstly to reduce the impact of low resolution on the shape integrity calculation;The k-means clustering algorithm is used to segment the image of the marking point into four regions.The complete contour of the target region of the highest luminance is extracted.Taking this as a reference,the luminance of different regions are weighted by using the segmentation of marking points and the shape integrity of the marking point is calculated.Finally,marking points are graded and assessed according to their shape integrity.