Object Tracking in Video Sequences

Praćenje objekata u videozapisima je proces detekcije i identifikacije jednog ili više pokretnih objekata u vremenu i važna je primjena računalnog vida. U okviru ovog završnog rada implementirano je praćenje temeljeno na algoritmu mean-shift. Za rad algoritma bilo je potrebno podatke iz videozapisa pretvoriti u algoritmu razumljiv oblik. U tu svrhu izrađen je histogram praćenog objekta te je isti proveden kroz histogram backprojection postupak kojim iz početnog okvira videozapisa nastaje razdioba vjerojatnosti položaja objekta. Algoritam je zatim testiran na videozapisima iz unaprijed pripremljene baze videozapisa te su rezultati testiranja uvršteni u ovaj rad.Tracking objects in videos is the process of detecting and identifying one or more moving objects in time and an important application of computer vision. Within this undergraduate thesis a mean-shift tracking algorithm was implemented. In order for the algorithm to work, it was necessary to convert the data from the video into an algorithm-understandable form. For this purpose, a histogram of the tracked object was created and it was passed to a histogram backprojection procedure that generates the probability distribution of the objects position from the initial frame of the video. The algorithm was then tested on videos from a predefined video database and the test results are included in this thesis