Article Preview
Top1. Introduction
The process of locating moving objects over time by using the camera in video sequences can be the definition for object tracking. Being a promising field, object tracking in video in many areas including surveillance(Ali & Dailey, 2012; Lee et al., 2014; Li et al., 2011; Nie et al., 2012; Rasid & Suandi, 2010) has been applied, robots navigation(Kaur et al., 2019; Rautaray & Agrawal, 2012; Tao & Yu, 2012) human r- computer interaction (Benavidez& Jamshidi, 2011; Capi et al., 2010) augmented reality and virtual reality (Biswas et al., 2014), GPS technology (Margulis & Galli, 2018), road-mapping (Daim et al., 2012) and three-dimensional (3D) computer graphics technologies (Elloumi et al., 2017).
However, modern technology has been seeing various challenges incorporating sudden object motion, shifting of objects' appearance, deformed object structures, and occlusion of the tracked object. Object tracking occlusion takes place in three forms: occlusion by background, self-occlusion, and inter-object.
When a tracked object occluded by a structure in the background occlusion by the background happens. In most cases, self-occlusion happens when one part of the object under tracking occludes another, while inter-object occlusion resulted by the occlusion of two tracked objects.
Tracking methods' default remedy to occlusion in the modeling of the motion of an object by linear and non-linear dynamic models, using the derived models for the constant prediction of the object location in case of occlusion, pending the reappearance of the object. Other techniques have also been used to address occlusion including silhouette projections, color histogram, and optical flow (Luo et al., 2019.).
Results of previous experiments on single object tracking in various trackers occlusion situations utilizing techniques such as Kalman filter, Particle filter, and Mean Shift will be presented and various occlusion-handling methods will be reviewed. This paper aims to investigate in-depth occlusion in object tracking and review some of the handling methods, categorize them, and identify the new methods devised to handle occlusion.
Furthermore, some crucial issues relevant to occlusion handling will be discussed, among them the use of an appropriate selection of motion models, image features and use of multiple cameras. The target of many object tracking applications is the ability to detect and track moving objects at a specific location automatically (Elgammal et al., 2002), which requires the performance of some low-level computer vision tasks including object tracking, occlusion handling, and unusual motion detection.
Object tracking can be defined as the process of monitoring spatial and temporal changes through a video sequence, including its position, presence, shape, and size (Ali & Mirza, 2006; Lee et al., 2014). On tracking objects, a lot of challenges still remain big, because of applying changes in appearance patterns of objects, unexpected object motion and the scene (Lee et al., 2014), object structures that are non-solid and most significantly are handling occlusion of tracked object (Lee et al., 2014).
The main purpose of this paper is to discuss some of the tracking algorithms and their results to compare between the methods used in these algorithms to handle tracking challenges especially the occlusion and the advantages and disadvantages for each method. Also, it presents some of the datasets that are used in these algorithms. Also, a proposed framework will be presented to improve the result of the tracking algorithm and handling occlusion with keeping real-time processing speed.