AUTOMATIC COUNTING PEOPLE BY USING COMPUTER VISION TECHNIQUES

Diplomsko delo se ukvarja s prepoznavanjem oseb, ki se gibajo, in njihovim štetjem. V sklopu diplomskega dela se je razvil algoritem za avtomatsko štetje oseb na osnovi video posnetka. Algoritem deluje tako, da na slikah, ki jih zajame s spletno kamero, poišče osebe. Za iskanje oseb se uporabi povprečna slika, ki se odšteje od zajete slike, nato pa se s pomočjo računalniškega vida ugotovi, ali je na sliki oseba. Na osnovi obdelane slike algoritem označi zaznano osebo na sliki, hkrati pa ustrezno spremeni števec za štetje oseb. Algoritem se je med izdelavo dela integriral v prototipno aplikacijo, ki s pomočjo grafičnega uporabniškega vmesnika prikazuje rezultate. Algoritem je bil testiran na desetih video posnetkih, na katerih so se odvijali različni scenariji vstopanja in/ali izstopanja oseb v/iz prostora. Na podlagi pridobljenih rezultatov je izračunana natančnost algoritma, ki je 80 %, hitrost obdelave posamezne slike pa je 0,125 sekunde na tipični današnji strojni opremi.The thesis deals with detection and counting of people in motion. A new algorithm for counting people on video has been developed. The algorithm works with images acquired by a webcam. It is based on an average image that is subtracted from the currently acquired image. The subtracted image is further analyzed by computer vision methods for person detection. When a person is detected, it is marked and the person counter is increased. The algorithm has been integrated into a prototype application with an integrated graphical user interface used to present the results. The algorithm has been tested on ten video sequences showing different scenarios of entering and/or leaving individuals into/from room. Detection accuracy of 80% has been obtained, with the average of 0.125 seconds required for the analysis of a single image on an up-to-date personal computer