| Popis: |
V magistrskem delu je predstavljena ideja izvedbe varovanja sodelujoče robotske celice, ki temelji na metodologiji varovanja prek nadzora hitrosti in gibanja. Podane in opisane so tudi druge standardne rešitve možnosti varovanja, ki pa imajo s praktičnega vidika določene pomankljivosti in omejitve. Te pomankljivosti so bile motiv za izdelavo univerzalnejše rešitve. Magistrsko delo nas vodi skozi postopek implementacije stereo vida, katerega namen je spremljati oddaljenosti objektov interesa, ki so ključni za zagotavljanje varnosti take aplikacije. Opisani so koraki ter tudi strojna in programska oprema, ki je potrebna za uspešno izvedbo aplikacije. Postopek se začne s kalibracijo posamičnih kamer zaradi distorzij, ki se pojavijo v procesu zajema slike, nato sledi stereo kalibracija z namenom lažjega iskanja oslonilnih točk med paroma slik in enostavnejše določitve razdalje objektov v prostoru. Za zaznavo objektov je bil uporabljen model nevronske mreže, ki se je iz baze učnih podatkov naučil detektirati želene objekte interesa. Kot zadnji korak sta predstavljena metoda iskanja oslonilnih točk na slikah in način izračuna razdalje objekta. V zaključku magistrskega dela so podani rezultati in kritična ocena aplikacije, pri čemer so opisane pomankljivosti in izzivi, ki so se pojavili med pisanjem tega dela. Na koncu so opisane tudi potencialne rešitve in možne izboljšave. The master thesis presents the idea of ensuring safety of a collaborative robotic cell, based on the speed and separation monitoring methodology. Also other standard safety approaches for a collaborative application are described, all of which have certain disadvantages and limitations. Those shortcomings were the motive for creating a more universal solution. The thesis guides us through the process of implementing stereo vision, with the intention of tracking the positions of our objects of interest, which is key to ensuring safety of such an aplication. All of the steps, as well as the hardware and software, necessary for its successful operation are explained and described. These steps begin with individual camera calibration due to distortion which appear in the image capture process. This is then followed by stereo calibration, which is needed in order to find the correspondng points between a pair of pictures more easily and also for distance estimation between objects in space. A neural network model was used to detect objects, which learnt how to detect the desired objects of interest from the learning database. Lastly, methods for finding corresponding points as well as object position estimation are described. The conclusion of the master thesis presents the results and critical appraisal of the application, outlining the shortcomings and challenges that emerged while writing the application. Finally, potential solutions and possible improvements are described. |
