A review of Features Detection Methods for recognition the location of a specific frame from Video Stream

Abstract

The recognition of an image is vastly employed in various application zones likerecognition of a shape, recognition of eye and recognition of a gesture. Among presentalgorithms for this objective, vision_based and statistical algorithms are further efficientsolutions for being precise, versatile and inexpensive. Vision_based algorithms can mostly beclassified as feature_based algorithms and appearance_based algorithms. In this paper, we canbe surveyed algorithms for recognition the location of a specific frame from video stream basedon both vision_based and statistical algorithms. To compare the performances of thealgorithms, we conducted a series of experiments on four types of algorithms such as HarrisCorner Detector, moment's invariants, Fast Retina Key-points (FREAK) Detector andhistogram matching. The comparison process relied on two important measures: time it takesto locate the image in the video stream and the accuracy. Experimental results on a sample ofvideos of different sizes showed that the time taken to detect the location of the image withinthe video stream is high when using the FREAK detection algorithm, while the time is verylittle when using the moment's invariants algorithm. As for the detection accuracy, it is highwhen using the FREAK detection algorithm, while the very little time when using the moment'sinvariants algorithm.