Article Preview
TopThe tracking of an object in an images sequence consists of estimating the position of this object in each frame of the sequence. The tracked object can be rigid or non-rigid, simple or articulated. During the tracking, several difficulties are encountered, some related to the tracked object: rotation, sudden/unpredictable movement, or partial/total occlusion. Other difficulties are related to the conditions of the scene: change in the scene illumination, variable background or degradation of the images quality due to the nature of the sequences or to an introduced noise.
Object tracking is realised in two phases (for each frame of the sequence): the first step is to search similar objects to the target, and then, choose the most likely one according to a similarity measure as, the cross-correlation (Bifulco, Cesarelli, Allen, Sansone, & Bracale, 2001), the sum of squared differences (Hager, Dewan, & Stewart, 2004), the Bhattacharyya coefficient (Shirinzadeh, Seyedarabi, & Aghagolzadeh, 2012), mutual information (Panin & Knoll, 2008) or correlation coefficient (Qin & Pun, 2018).