Comparison of different integral histogram based tracking algorithms

Abstract

Object tracking is an important subject in computer vision with a wide range of applications – security and surveillance, motion-based recognition, driver assistance systems, and human-computer interaction. The proliferation of high-powered computers, the availability of high quality and inexpensive video cameras, and the increasing need for automated video analysis have generated a great deal of interest in object tracking algorithms. Tracking is usually performed in the context of high-level applications that require the location and/or shape of the object in every frame. Research is being conducted in the development of object tracking algorithms over decades and a number of approaches have been proposed. These approaches differ from each other in object representation, feature selection, and modeling the shape and appearance of the object. Histogram-based tracking has been proved to be an efficient approach in many applications. Integral histogram is a novel method which allows the extraction of histograms of multiple rectangular regions in an image in a very efficient manner. A number of algorithms have used this function in their approaches in the recent years, which made an attempt to use the integral histogram in a more efficient manner. In this paper different algorithms which used this method as a part of their tracking function, are evaluated by comparing their tracking results and an effort is made to modify some of the algorithms for better performance. The sequences used for the tracking experiments are of gray scale (non-colored) and have significant shape and appearance variations for evaluating the performance of the algorithms. Extensive experimental results on these challenging sequences are presented, which demonstrate the tracking abilities of these algorithms.