Abstract:
Quality control of textile fabrics is a crucial problem for textile Industries. The inspection of fabric is done manually by experts which is a hectic and eye straining procedure. The conveyor belt of fabric machine is around 3m wide and moves with a speed of around 25 to 200 m/min. Humans cannot detect more than 60% defects and there is chance of error even at the speed of 25m/min. Therefore, an automated Intelligent system is required efficient both in terms of precision and speed, for Inspection of fabrics. The system must have both high accuracy and good frame rate. Current systems have either high Accuracy or high FPS (less processing time) since there is a trade-off between both. Machine Learning is computationally and economically very expensive to detect and mark faults on the images. Several existing vison-based systems are also either specific to certain type of defects or dependent on color of fabric, such systems have either low accuracy or low FPS, hence yielding poor results or very expensive for small companies.
We have designed a very simple yet very fast system with trivial pixel processing algorithm, which is optimized, is far more accurate than trivially running systems and inexpensive approach towards the resolution of such problems. Through our image acquisition system, we have acquired very High Definition Images and have processed them by our implemented system at very high frame rate on trivial personal computers. The approach works on all colors and for all types of defects with the same accuracy, gives coordinates of the fault to the robotic arm which marks the faults/cuts the defective area/raises alarm etc. A data set of 107000 images is used having dimensions of 1920x1080. Experiments yielded the results of over 93% on images and up to 100 % on video.
