The Research On Real-time Online Vision System For Fabric Defect Detection

In the rapid development of modern textile industry, quality control has become a industry focus. However, the low efficiency, poor accuracy, high cost human vision inspection approach has occupied the dominant position in practical production, which will no longer meet the need of modern textile industry. Automatic fabric defect detection will be a trend in the future. This research concerns how to apply machine vision technique to practical fabric defect detection field and proposes a low cost, high performance and excellent expansibility solution, which is meaningful for automatic inspection in fabric defect detection.The detection system was introduced from three aspects of hardware platform, image processing algorithm and application software. Hardware platform is mainly consisted of mechanical structure, transmission and image acquisition subsystem, each part has its components selection, installation and debugging. Image processing algorithm has four processing parts including image acquisition, pre-processing, image segmentation and post-processing. In these parts, double buffer alternative acquisition technique is applied to image acquisition to improve real time performance; the main task of pre-processing is to do image smoothing, enhancement and stitching to improve image quality; image segmentation adopts a concept of multi-algorithm combined approach, which regards each defect segmentation algorithm as a single detection unit integrated in software system. Then different detection units are employed at different fabrics and defects to gain better detection efficacy; Post-processing responsible for detection results localization and storage. Application acts the role of bridge between hardware and algorithm. It makes the system more practicable for the reason of its online detection, user management and other powerful functions.Finally, through various online-offline experiments, all the performance criteria were validated and reach the expected goal. In addition, system application to industry field was analyzed and summary of research weakness has also been done.