Person Tracking in Video Sequences

Zbog primjenjivosti u stvarnom životu, praćenje osoba kao segment računalnog vida je iznimno zanimljivo. Kroz ovaj rad prikazani su neki od algoritama putem kojih se može ostvariti tražena funkcionalnost. Programsko ostvarenje je putem programskog paketa OpenCV, a ostvareno je u jeziku Python. Korišten je Python 2.7 i OpenCV 3.1. Podržani algoritmi praćenja su: Boosting Tracker, MIL Tracker, KCF Tracker, TLD Tracker i Median Flow Tracker. Korisniku je omogućeno označavanje jedne osobe unutar videozapisa putem pravokutnog okvira. Prema poziciji okvira, dalje se koriste navedeni algoritmi putem kojih se okvir pomiče po području kretanja osobe unutar videozapisa. Kvaliteta praćenja ovisi i o početnom odabiru okvira, kao i o korištenom algoritmu. Funkcionalnost svakog od navedenih algoritama povezana je s videozapisom na kojem se određeni algoritam koristi. Zbog toga je pri odabiru algoritma bitno poznavati njegove karakteristike, kao i video na kojem će se primjenjivati.Because of its applicability in real life, tracking as a computer vision segment is an area under constant development. This paper presents some of the algorithms that can be used to achieve the required functionality. The program is realized through the program package OpenCV, and it is implemented in the programming language Python. Versions that were used are Python 2.7 and OpenCV 3.1.Implemented tracking algorithms are: BOOSTING Tracker, MIL Tracker, KCF Tracker, TLD Tracker and Media Flow Tracker. The user is enabled to tag one person within the video through the rectangle area called bounding boxes. The algorithms listed above are used to move the bounding box around the person's motion area within the video. The quality of monitoring also depends on the initial selection of the frame, as well as on the algorithm that is being used. Therefore, when selecting the algorithm which is going to be used, it is essential to know its characteristics as well as the video to which it will be applied