Research On Nondestructive Tire Defect Detection Using Computer Vision Methods

Nondestructive tire defect detection has been becoming a significant research field due to its important role in traffic safety, tire industry and the development of a retreaded tire industry. Different methods are proposed by researchers on tire defect detection which accelerate the applications of computer vision based methods in tire industry. The basic research questions related to this problem including:image processing, image analysis, texture image segmentation and pattern classification etc. which are very important research fields in computer vision with application to industrial defect detection, medical images analysis, remote sensing image analysis and biometric identification and many other fields.In this work, image processing, image analysis and pattern classification techniques are used to research on the key problems of tire automatic nondestructive defect detection based on computer vision. Image enhancement, image segmentation, edge detection, wavelet multi-scale analysis and pattern classification methods are used on tire laser shearography images and X-ray images to detect defect information. An architecture of tire tire automatic nondestructive defect detection system is proposed. Experiments are designed to validate the performance of the proposed scheme. The main creative contributions of the work are as follows:1. A tire X-ray image segmentation method is proposed based on texture analysis. Optimal Gabor filter parameters are selected through experiments which would contribute to Gabor filters design for the target images. A fuzzy c-means clustering algorithm is used to cluster Gabor feature images so that tire X-ray image is segmented. Experimental results indicate that regions of different textures are segmented clearly and precisely. It will be easier to design defect detection algorithm for segmented images since the background texture is uniform.2. A tire X-ray image defect detection method is proposed based on total variation image decomposition. A tire X-ray image can be decomposed using total variation into texture part and cartoon part which represent tire sidewall cord and tire sidewall rubber respectively. Therefore, we can detect tire sidewall rubber defects (such as bubbles and foreign objects etc.) and cord defects separately using more targeted algorithms easily.3. A tire sidewall rubber defect detection approach is proposed based on curvelet image enhancement and improved Canny edge detector. Due to the capability of line singularity representation, defect edges are enhanced nonlinearly using curvelet. Thereafter, defect edges are detected using improved Canny edge detector. This would result in a reconstructed image more convenient for edge detection and the time complexity is reduced on the other hand. Furthermore, the eight-neighborhood bilinear interpolation non-maximum suppression method is introduced to improve the performance of Canny edge detection. Our detection results are evaluated on test images using the proposed scheme and compare favorably to the state-of-the-art methods.4. For tire sidewall cord defect, a statistical based method is proposed. Three expert rules are designed to discriminate if a cord is defective or not using cord statistical information. Experimental results show that the proposed scheme can effectively detect defects in the tire sidewall.5. A tire defect detection scheme based on wavelet multi-scale analysis and morphology operation is proposed. We address the problem of tire defects characterization in ways of local regularity analysis of signals. Optimal scale and threshold parameters are given based on experimental analysis and discussions. Mathematical morphology operation is utilized to remove residual background texture edges and noise. Comparisons with conventional image edge detection methods show that our scheme outperforms these methods in detecting defect edges. The proposed scheme has been evaluated on the test dataset.6. A tire defect classification method is proposed based on support vector machine (SVM). A compact feature space is designed using shape features, gray level features and gray level co-occurrence matrix based secondary statistics. Different kernels are discussed and RBF is selected as kernel function. Optimal penalty parameter C and kernel parameter a are selected by cross-validation. Feasibility and validation is verified through experiments.