Semi-automatic ground-truth trajectory extraction on image sequences

Karabıyık, Murat
In this thesis, offline semi-automatic ground-truth trajectory extraction technique is proposed that uses measurements of detector as basis. The unknown camera motion of the videos used throughout the thesis makes the problem even more challenging. The camera motion is estimated by using a novel method which uses a special Kalman filter. Background objects are discriminated from the targets and they are used to estimate the camera motion. Two different trackers are implemented to extract the ground-truth. Measurements of the detector are tracked by using Tracker-1. The tracks resulted from Tracker-1 are associated by using Tracker-2. The velocity difference between the target and the camera is used both for position predictions of Tracker-2. The user of the program gives the true target information for the first frame. The output of Tracker-2 gives the raw ground-truth and it is smoothed via Kalman smoother. The output of the Kalman smoother gives the ground-truth. Finally, an example tracker which is used in real time tracking problems is evaluated by comparing the ground-truth and measurements of the tracker which is evaluated. 
Citation Formats
M. Karabıyık, “Semi-automatic ground-truth trajectory extraction on image sequences,” M.S. - Master of Science, Middle East Technical University, 2017.