دانلود مقاله ترجمه شده ردیابی هدف تنها با نشانه پس زمینه – مجله IEEE

• فهرست مطالب:

 چکیده
۱ مقدمه
۲ کار مربوطه
۳ ردیابی تنها با نشانه های زمینه
الف ردیابی چهار چوب
ب نقشه اعتماد زمینه
ج حرکت مدل
۴ نتایج تجربی
الف نصب تجربی
ب نتایج در سکانس های زمینه ایستا
ج نمایش نتایج در سکانس های زمینه پویا
۵ نتیجه گیری

• بخشی از ترجمه:

در روش های قبلی برای ردیابی شی، نشانه زمینه معمولا به عنوان مکمل و یا همراه با نشانه هدف استفاده می شد. آنها می توانند به طور مستقل مورد استفاده قرار گیرند. دراین مقاله، ما برای اولین بار ردیابی شی را با استفاده از نشانه های زمینه مطالعه می کنیم. یک روش جدید برای ردیابی شی با جایگزین کردن شباهت با معیار هدف مورد استفاده روش های قبلی ارائه شده است. نتایج تجربی نشان دهنده اثر بخشی رویکرد ما است. توجه داشته باشید که این مقاله استفاده از روش ردیابی پیشنهاد شده نسیت اما پدیده را نشان می دهد. پیگیری یک شی بدون استفاده از شکل ظاهری آن به نظر باورنکردنی است، اما نتیاج تجربی گسترده ما به وضوح نشان می دهد که در واقع آن عملی برای هر دو سناریو زمینه ایستا و پویا است. این ثابت می کند که نشانه های زمینه می تواند به طور مستقل و تنها در ردیابی شی استفاده می شود.
این پدیده یک مرجع جدید در طراحی روش های ردیابی را فراهم می کند. در روش ردیابی تبعیض فعلی، اثر متقابل بین اطلاعات هدف و زمینه منجر به پیش بینی بهتر می شود. با این حال، ترکیب آها با هم همیشه بهتر از استفاده یکی از انها نیست اگر دیگری به شدت آلوده باشد. نتایج ما نشان می دهد که تشخیص آلودگی و به طور موقت از بین بردن یکی از آلوده کننده ها ممکن است یک انتخاب بهتر از استفاده هر دو باشد. حذف نه تنها می تواند اطلاعات زمینه، بلکه نشانه های هدف نیز باشد.

• بخشی از مقاله انگلیسی:

INTRODUCTION OBJECT tracking is a fundamental problem in computer vision [1]. Great progress has been made in constrained settings, such as the scenarios with rigid objects or stationary cameras. Although lots of approaches have been proposed during the past decades, for unconstrained scenarios, robust tracking still remains a challenging and unsolved problem. One of the key challenges is the appearance variation of the object. In real-world scenarios, variations in the appearance of an object can be caused by many factors, such as shape deformation, pose variation, and occlusion. Different factors usually cause different variations, and the appearances of different kinds of objects may also vary in different ways. Therefore, it is very difficult to handle all kinds of variations with one unique solution due to the great variety of possible changes. In [2], several state-of-the-art methods are evaluated on sequences with different challenging factors. The experimental results show that none of these methods can perform well on all the sequences. How to cover more kinds of variations is a complicated and difficult problem. In the past decades, many object tracking methods have been proposed [1]. However, the above-mentioned problem still has not been well addressed. In essence, a better tracker requires a higher level of understanding about the visual objects, which is quite complex and challenging. In this paper, we show that this may be avoided in some scenarios. Different from the previous approaches that focus on the objects, we study the object tracking problem only using background information. Our study is based on the following observation. In an image sequence shown in Fig. 1(a), a moving object is replaced by patches of random colors. By observing this sequence, we can perceive the movement of the object as well as observe the original sequence shown in Fig. 1(b). The appearance consistency of the object is the basis of most object tracking approaches. Although it does not hold in the sequence (a), we can still perceive the movement. This implies that other cues may help us to perceive the movement. In the sequence (a), the patch color is not consistent, but it is always different from the background color. In other words, we can perceive the movement not only when it is the object but also when it is not the background. Compared with the being-the-target criterion, the not-beingthe-background criterion possesses a natural advantage, i.e., robustness to the appearance variation of an object. When an object changes its appearance, the difference between the object and the background may not necessarily shrink. Thus, the problem of appearance variation can be considerably avoided. The above observation and analysis lead to an interesting question: can an object be tracked using only background cues? We study this question in this paper and show the feasibility of tracking visual objects using only background cues. Background cues have been used in the previous studies of object tracking. For the scenarios with stationary cameras, background subtraction is proved to be effective [1]. For the approaches designed for dynamic cameras, background cues are combined with target object cues. However, background cues usually play a supplementary or companying role, which means that background information cannot be used independently in these approaches. Different from the previous approaches, background cues are used independently for object tracking in this paper, which is proved to be feasible according to the extensive experimental results. The remainder of this paper is organized as follows. Section II gives a brief review of the related work. The proposed tracking approach based on background cues only is described in Section III. Section IV shows the experimental results. Conclusion and discussion are given in Section V. II. RELATED WORK Object tracking is an active topic in the computer vision literature. A number of methods have been proposed during the past three decades. From the perspective of background information utilization, the previous object tracking approaches can be divided into two categories. Methods in the first category usually construct trackers by describing the target itself [3], [4], while those in the second category use the information of both the target and the background.