System for detecting people in video sequences

Detekcija osoba zahtjevan je posao zbog raznolikosti pojavljivanja osoba na slikama i videima. U ovom radu se opisuje pristup detekciji osoba algoritmom Viole i Jonesa koji funkcionira kvalitetnim radom grupe slabih klasifikatora koristeći Haar značajke i AdaBoost. Opisuje se prikupljanje, obrada i korištenje slika za treniranje detektora, pokušaji treniranja i statistika uspješnosti. Prikazani su primjeri raznih detekcija u slikama i videima i opis načina postizanja dobivenih rezultata. Za izradu je korišten programski jezik Python i naredbe opencv_createsamples i opencv_traincascede iz OpenCV biblioteke za računalni vid kao i skripte za detekciju i pripremu treninga.People detection is challenging job because of big variety of people appearing in pictures and videos. This document describes the approach to people detection by Viola and Jones algorithm which works by good group work of the weak classifiers using Haar features and AdaBoost. Collection, processing and usage of images to train detectors, training attempts and performance statistics are described. Examples of various detections in images and videos and way of achieving results are shown. Python programming language and commands like opencv_createsamples and opencv_traincascade, which are part of OpenCV library for computer vision, are used for creation as well as scripts for detection and preparation of training