Detection of free parking places in a reconstructed point cloud

Razmatra se autonomni sustav za pronalazak slobodnog parkirnog mjesta u rekonstruiranom oblaku točaka. Opisan je stereoskopski vid te su objašnjene metode rekonstrukcije oblaka točaka u monokularnom videu. Razvijeni sustav na ulaz prima rekonstruirani oblak točaka dobiven iz kamera na vozilu. Iz oblaka točaka najprije se utvrđuju najbliže prepreke oko vozila. Zatim se pristupa metodi određivanja orijentacije prepreka (u kojem se smjeru proteže parking). Sljedeći korak je utvrđivanje bližeg i daljnjeg ruba parkinga pomoću histograma udaljenosti svih najbližih prepreka. Zadnji korak je utvrđivanje slobodnog prostora za parking nakon što su poznate koordinate i udaljenosti rubova parkinga. Ispitni skup sastoji se od 9 scenarija od kojih je 4 napravljeno u kontroliranim uvjetima, a 5 u stvarnoj vožnji. Pokazuje se da sustav vrlo dobro funkcionira ako je rekonstrukcija oblaka točaka zadovoljavajuća, odnosno bez prevelikog šuma.An autonomous system for finding a free parking space in a reconstructed point cloud is considered. A stereoscopic view is described and the methods of reconstructing point cloud in the monocular video are explained. The developed system receives reconstructed point cloud from the cameras on the vehicle. Firstly, the nearest obstacles around the vehicle are determined. Next is determining the orientation of the obstacles (in which direction is the parking). The next step is to determine the nearest and further edge of the parking lot by the distance histogram of all closest obstacles. The last step is to determine the free space for parking after known coordinates and distance of the parking edges. The test set consists of 9 scenarios, of which 4 were made under controlled conditions and 5 in the real driving. It is shown that the system works very well if the point cloud reconstruction is satisfactory without too much noise