32 Schneider, Dorian Merhof, Dorit Vary, Peter 611710 On-loom fabric defect detection : state-of-the-art and beyond Publikationsserver der RWTH Aachen University Aachen English XI, 213 S. : Ill., graph. Darst. Weaving is one of mankind’s oldest crafts. The process of interlacing two sets of yarns in an orthogonal way according a predefined pattern is a technology which is as old as human civilization. Over the centuries, the textile industry evolved into a high-tech industry, characterized by highly sophisticated production machines which operate mostly autonomously and are uncoupled from any human interaction. Built into safety relevant products like airbags, safety belts, fire resistant clothing, bullet-proof cloth or artificial vascular grafts, technical woven fabrics impose highest production quality standards. Reliable and fast quality assurance is thus crucial. The industrial standard approach for quality assurance is still based on human, manual inspection. In order to augment the production throughput, to achieve lower labor costs and to guarantee stable product quality, the development of reliable methods for fully automatic fabric quality control has become a vital topic for research around the world. Within this thesis, the development of a novel, loom-integrated automated visual inspection system for high resolution woven fabric defect detection is described. Accordingly, this work is divided into three major parts. Part I investigates a set of 14 selected state-of-the-art fabric defect detection algorithms to assess the current detection performance of existing methods. The study is conducted with unified fabric image databases and assessment metrics and therefore represents the first fabric defect detection benchmark of this kind published in literature. Motivated by the benchmark results, Part II discusses the design of a novel, high-resolution traversing camera inspection system for locating and classifying potential defects in woven fabrics. The implemented prototype can track single yarns in real-time during production and measures fabrics with regards to geometry, extent, orientation and shape for the first time. This is a major improvement compared to hither to approaches that treat fabrics as near regular texture and apply pattern analysis algorithms to detect defects. The detailed description of the image processing pipeline is complemented by a comprehensive on-line evaluation and in-depth discussions about mechanical system integration, vibration damping, imaging strategies and product costs. The proposed image processing framework is finally extended in Part III by two additional algorithmic features. First, a method is discussed that allows an automatic classification of fabric weave patterns without any prior knowledge about the investigated material. Furthermore, an algorithm for adaptive measurement of changing yarn densities is presented. Again, both extensions were extensively evaluated and the results are directly compared to state-of-the-art performance measures. Aachen, Techn. Hochsch., Diss., 2015 ; Dissertation / PhD Thesis Dissertation Aachen, Techn. Hochsch. 2015 RWTH-2015-04739 2015 https://publications.rwth-aachen.de/record/483740