Multiple Object Tracking Using SIFT Features and Location Matching
Seok-Wun Ha, Yong Ho Moon · 2011
In recent, object recognition and tracking systems have been developed that use local invariant features from Shift Invariant Feature Transform algorithm. Most of them are implemented by distance matching of descriptor features between the reference and the next consecutive frame image. Among the matched keypoints generated from SIFT descriptor matching, there are some mismatched keypoints when keypoint location information is considered. These location-mismatched keypoints could be affected to object tracking perfomance. To achieve a stable tracking it is necessary that these are detected and discarded in tracking action. In this paper a robust object tracking system is presented that strengthen stability in tracking by eleminating location-mismatched keypoints. Experimental results show that a stable and robust tracking can be achieved in a test video sequence includes multiple objects.