Computer vision automation system for sorting partially overlapping tiles

Traditionally, a method of manual sorting of tiles based on color is being performed by human operators via visual inspection. This method is slow and tedious. Another automatic method has been developed using assembly line machines, but it requires a significant amount of space to operate. New automatic tiles sorting method based on color using a robotic arm is proposed in this study which is more effective and does not require large physical space. This method utilizes machine vision prior to sorting, however it also faces several challenges. One of these challenges is to differentiate between similar color tiles which are partially overlapped. Another is to distinguish between white tiles and the back of overturned white. The aim of this thesis is to develop Color-based Automatic Tiles Sorting system (CbATS) to mitigate the mentioned challenges. The CbATS consists of three main components which are a color-detection algorithm for distinguishing tiles according to the color, image segmentation that ensures the separation between partially overlapped tiles, and texture features extraction method to determine overturned tiles. For the first component, three color-based models were implemented and compared. These models are Hue, Saturation and Value (HSV); Red, Green, and Blue (RGB); and Luma (brightness), Blue-difference, red-difference chroma components (YUV). The color models are employed to investigate the effectiveness of differentiating tiles based on color. The Watershed Distance Transform with H- minima (WDTH- minima) is utilized in the second component with different H- minima to produce sufficient separation results for partially overlapped tiles. A texture feature extraction algorithm based on (standard deviation of intensities and entropy) were developed and compared in the third component to identify overturned tiles from white tiles. The results show that color detection using HSV model produces 100% accuracy when a yellow light is used. Besides that, using WDTH-minima segmentation method with (H-minima<=1) produced 100% of accuracy for separation tiles. Furthermore, calculating standard deviation to determine the texture feature, obtains 100% of accuracy in distinguishing between ''white tile'' and overturned tiles. Combination of the methods HSV, WDTH-minima, and standard deviation significantly improved the accuracy of sorting that reached 100% for the overall proposed system.

Text FK 2020 28 - ir.pdf Download (2MB)

Actions (login required)

View Item