Industrial partner: MAS Active Trading Pte. Ltd
The Stitches Per Inches (SPI) is a key quality figure in apparel industry
Different speed of the machinists, variations in machine settings hinder getting SPI within acceptable limits.
The usual practice is to find SPI manually on some randomly selected parts of the garments which consume time and effort.
The prototype device was implemented to automate counting system as applied to 6 different seam types and object count varies in the range of 16 – 24.The execution time of the device is reduced to 6 seconds whereas it consumes around 40 -60 seconds for the manual operation.
The approach used in the prototype was:
o A still image of the fabric sample was acquired using a digital camera and it is then converted to a grayscale image.
o As both background and foreground are in same color, texture information was used to segment the object from the overall image.
o The foreground has rough edges that can be visualized as a function of intensity variation.
o Sobel edge detector was adopted in the proposed methodology for every seam types.
o The still image was blurred before applying Sobel mask and it is then converted to a clear binary image by defining a threshold value.
o The binary image was compressed into a vector where peaks correspond to the object count.
o Several images were subjected to the algorithm and averaging the result yields an accurate representation of stitch count.
Object Counting Algorithm to Count Stitches on Fabric