Research On Rubber Tread Quality Detection Based On RGB-D Camer

With the rapid development of the rubber industry,companies are placing higher demands on the quality control of rubber products and intelligent inspection of production lines.Existing domestic inspection links often rely on manual or weighing correlations,so the application of reliable machine vision technology is the key to effectively reducing inspection costs,improving inspection accuracy and safeguarding tyre quality.Based on RGB-Depth camera with continuous improvement of depth image acquisition quality and analysis of tread quality characteristics,this paper studies the application of RGB-D visual technology in product quality inspection,and realizes the measurement and positioning of tread size and surface defects after length cutting.The main work of this paper is as follows:(1)The camera calibration of the Astra pro depth camera was completed.For the influence of infrared projection speckle,a mathematical model of corner and reprojection error is established.Non-dominated Sorting Genetic Algorithm-II is used to filter corners with large error,78% of the corners are used for checkerboard calibration experiments,the reprojection error and transformation matrix results are evaluated,and the internal parameters,transformation matrix and reprojection error of RGB,IR cameras that meet the requirements are determined.(2)Improved the traditional pixel metric ratio measurement method.In response to the impact of noise interference pollution,the Noise2 Noise unsupervised denoising algorithm based on the UNet network was used to remove noise.The Peak Signal to Noise Ratio value was used as an evaluation indicator to compare the effectiveness of different denoising algorithms.After five sets of noise denoising experiments,good denoising effects were achieved.On this basis,combined with the principle of pixel metric ratio,the Canny edge detection algorithm is used to locate the tire edge at the pixel level,measure the size and length of the tire profile,and compare the image detection experimental results of the denoised and original clean images before and after denoising.The effectiveness of the improved method in unsupervised denoising is verified,and the size detection results of the denoised and clean images are basically consistent.(3)The method of tread size and surface defects detection based on point cloud data was proposed.To overcome the shortage of scale conversion accuracy in pixel metrics ratio,Random sample consensus and Oriented Bounding Box are introduced to measure the point cloud data on the tread surface,fit the transfer datum plane coordinate system,and generate a surface compact boundary box.The precise detection of the tread size is completed,and the absolute and compact bounding boxes of five groups of experiments are compared.Relative error data show that the detection method has a precision of 1mm and a measurement deviation of 0.14-2.67%,which meets the highest accuracy standard of the national standard.For the processing of defective point cloud data,the Iterative Closest Point algorithm is used to accurately register the two-sided point cloud,filter the point cloud data which is higher than the Euclidean distance mean between the point clouds,and set it as defective point cloud.The experimental results show that the defective location of the tread can be visually displayed.In conclusion,the RGB-D visual inspection method constructed in this paper is a potential solution which can enhance pick-up guidance and perform tread quality detection because of its low cost and high detection accuracy.