Detekce objektů náhodně inicializovaným sledováním

Abstract

Detection and tracking objects appearing in camera data is one of a main topic in computer vision. In many systems the fast camera data processing is desired. In this diploma thesis the fast algorithm for tracking rigid parts of known objects by sequences of learned linear predictors was implemented. We purpose use the same algorithm also for detection know object in picture. The principal drawback of tracking by linear predictors in fast motion (failure by reason of limited range of learned predictor) was supress by redetection in neighbourhood of the last know object position. The real objects was tracked by algorithm in real sequence of pictures. Algorithm was also used for detection human face in picture. The ability of algorithm for tracking and detection (achieved framerate and effectivity of detection and redetection) was verified by experients.

Movies:

• face_tracking [video]

• mouse_pad_tracking [video]

• some detected faces [zip-archive]