Search

Diplomska naloga obravnava tematiko avtomatiziranega pobiranja izdelkov v razsutem stanju. Problem, katerega rešitve so predstavljene tekom naloge, je prepoznava lege in orientacije objektov, ki so naključno razporejeni v prostoru. V prvem delu naloge je opravljen pregled obstoječega stanja tehnike na področju pobiranja izdelkov in tridimenzionalnih merilnikov ter povzeto teoretično ozadje nekaterih uporabljenih funkcij za prepoznavo izdelkov. V drugem delu diplomske naloge smo uporabili 3D merilnik za zajem 3D slik objektov v razsutem stanju. Sledil je razvoj programske opreme v programskem jeziku Python, s katero smo nato obdelali 3D slike in jih pretvorili v intenzitetne globinske slike. V nadaljevanju smo te slike še dodatno obdelali z različnimi funkcijami, ki so dostopne znotraj odprtokodnih knjižnic ter jih nato uporabili kot vhodne podatke v treh različnih funkcijah za prepoznavo enostavnih objektov. V zaključku naloge smo preverili delovanje posamezne funkcije na različnih scenskih slikah, na katerih so bili objekti različnih oblik. Na podlagi rezultatov smo ovrednotili posamezno funkcijo ter ocenili, ali je primerna za prepoznavo izdelkov v razsutem stanju. The final thesis deals with the topic of automated collection of bulk products. The problem, whose solutions are presented during the task, is to recognize the position and orientation of objects randomly distributed in space. The first part of the thesis gives an overview of the current state of the art in the field of product picking and three-dimensional scanners and summarizes the theoretical background of some of the functions for the identification of products. In the second part of the thesis a 3D scanner was used to capture 3D images of bulk objects. This was followed by the development of a software in the programming language Python, with which we processed the 3D images and converted them into images with depth of field. Later, we have further processed these images with various functions available within open source libraries, which we then used as input in three different functions to identify simple objects. At the end of the assignment, we checked the results of each function on different sets of stage images showing objects with different shapes. Based on the results, we evaluated each function and estimated whether it was suitable for identifying bulk products.

V zaključnem delu predstavimo razvoj in delovanje programa za iskanje, prepoznavo in razvrščanje krožnih oblik v stereo sliki. Podan bo pregled uporabljenih algoritmov in metod za reševanje tega problema, na kratko pa opisanih še nekaj ostalih metod s tega področja. Podrobneje so opisani posamezni koraki v postopkih načrtovanja in realizacije projekta. Na kratko je predstavljena uporabljena strojna oprema, ter njene prednosti in pomanjkljivosti. Naloga se v nadaljevanju osredotoča na programski vidik reševanja problema. S stereo kamere zajemamo 3D slike, s pomočjo katerih lahko izračunamo globinsko sliko prostora. Na dobljeni globinski sliki iščemo kroge, ki predstavljajo okrogle predmete v prostoru. Z uporabo samo ene slike lahko enak prizor analiziramo v 2D pogledu. Z iskanjem, obdelavo, primerjanjem robov in podatkov iz obeh pristopov lahko dobimo precej zanesljiv način zaznave krožnih oblik v sliki. The final work presents development and functionality of a program, designed to search for, recognize and categorize circular forms in a stereo image. An overview will be given about the algorithms and methods used in solving this problem, as well as a short summary on some of the other available methods used in this field. Steps taken during research, development and production of this project will be described in further detail. The final work gives a short presentation of the hardware used in this project, briefly explaining its advantages and disadvantages. Next, it focuses on the programing aspect of the solution. A stereo camera is used to capture 3D images, which we can use to calculate the depth field of a given scene. The resulting depth map is used to search for circles, which represent round objects in space. By using only one of the images captured, we can analyze the same scene in a 2D view. By looking for, processing, and detecting edges, as well as comparing data from both procedures, we can achieve a rather reliable method of detecting circles in a given image.