What's it all about:
Implementation of positional recognition of objects in automated assembly
This diploma thesis focuses on chances of using some computer techniques to recognize shapes of objects in automated assembly. First of all there's description of commonly used practics in this field. The core of my work here is design of edge detector that should be faster and more efficient than usually used edge detectors. At the same time this work focuses on proposition of automated assembly's machinery so new edge detector could be successfully used in real assembly of circuit breaker RI 60.Main focus is to make production more cost-less.
Program itself is written in Builder C++ 6 and is standalone Win32 application. It provides edge detecting procedures for images stored in .bmp format. Edge detectors that are implemented are: Laplace 1, 2, 3, Sobel-Prewitt and Prewitt-Robinson masks. Others detectors were designed and implemented by me, and I've decided to call them :
diferenciálne štvorsusedstvo (differential four-neighborhood)
prehľadávanie zľava do prava (left to right edge detection)
thresholding (doesn't work like edge detector, was implemented purely for fun)
As a matter of fact, those algorithms proven to be faster and more efficcient than traditional edge detecting masks. Output from those edge detector was supposedly to be used in neural network based processing unit
This output is simple text file with 1 where edge is located, 0 else where. This file is input to neural network where exact location of object is determined (not part of my diploma thesis).
As the time has passed, I found some minor problems in thesis:
Program has checkbox to determine whether final image (with edges detected) should be displayed or only text file is on output. Values printed in thesis are all with checkbox checked, so they are like 20-times greater than they really are.
There are some weird average times of experiments. However my algorithms are still faster.
Reference table for some images is out-of-date.