There are many object tracking algorithms using optical flow methods. Existing literature in flow estimation are abundant, however, there is not any method found to extract perfect flow data. So making an optimal choice of the methods that is suitable for moving object tracking applications remains an open problem. The purpose of this Thesis is to provide a survey of existing optical flow techniques for object tracking applications. We experimented state of the art flow methods using several video clips of a static camera including indoor and outdoor scenes. Furthermore, we present a novel object tracking system for sparse and dense flow extraction methods by using inverse flow optimization technique. Sparse algorithm process only some pixels from whole image, which are easy to follow and dense optical flow algorithm can process all pixels in image. The both flow methods have some hard parameters in their implementations. In hard parameter coding the source code, have to be changed any time the input data or desired format changes. It means users have to carefully test and forethought the system code performance. The best way to set the hard parameters is cross validation over user desire dataset. In order to show the performance of the proposed tracking algorithm, some experiments have been performed under higher frame per second video clips. The results show that selection and cross validation of hard parameters for flow extraction methods are vital to get the best results.
展开▼
