Visual object tracking is increasing the interest of researchers in the field of computer vision. The applications of object tracking include video surveillance to ensure security and safety, people monitoring, traffic monitoring and human computer interaction. Background subtraction methods have received a lot of attraction due to less computation time and accurate detection of moving objects. However, difficulties in object tracking arise due to abrupt motion, changing the appearance of patterns for both the objects and the scene, non-rigid object structure, occlusions and camera motion. The goal of this paper is to review various object detection and object tracking methods using background subtraction. We categorize the background subtraction methods into five groups. We describe various methods in each group in detail and give a quantitative analysis of these methods.
Keywords: Background subtraction, Gaussian Mixture model (GMM), Fuzzy color histogram (FCH), Support vector machine (SVM)
1. INTRODUCTION
Video surveillance system have gained a lot of importance in the field of research because of availability of high end computers and increasing demand of video analysis. It has tremendous potential in various applications like securities for communities and important buildings, traffic monitoring in cities and expressway, detection of military targets, human computer interactions, weather forecasting etc. [1]. A typical video surveillance system involves intelligent features like motion detection, tracking of objects frame by frame, classification, recognition and synthesis of object. It also involves understanding and describing behavior of an object in the system. Motion detection is used in many computer vision applications like recognition of traffic situations, human actions, alarming in anomalous situations, monitoring the traffic flow and conjunction analysis, driver assistance and detection of his intentions, face detection, adaptive response to critical motion of nearby vehicles, human computer interaction, remote image processing, collision prediction of pedestrians etc. [3]. Actually tracking is nothing but to find a given object in the plane as it moves in the scene. It also gives other information about object like orientation, area or shape of the object [2].
1.1 Methods of visual object tracking
According to survey of various research [4] motion detection methods can be categorized into three major classes viz. temporal difference [5], [6],optical flow [7], [8], and background subtraction [9]-[11].In temporal difference method consecutive two or three frames are compared pixel by pixel basis. These pixel differences are used for calculating motion sequences. An adaptive threshold value is used to compute the temporal differences for detection. It is assumed that when there are moving objects, the image intensities are not varied in a short interval of time [12] so that motion can be detected by boundaries. These methods...