摘要
Background subtraction, although being a very well-established field, has required significant research efforts to tackle unsolved challenges and to accelerate the progress toward generalized moving object detection framework for real-time applications. The performance of subsequent steps in higher level video analytical tasks totally depends on the performance of background subtraction. Recent years have witnessed a remarkable performance of deep neural networks for background subtraction. The deep leaning has paved the way for improving background subtraction to counter the major challenges in this area. Also, the fusion of multiple features leads to the improvement of conventional background subtraction methods. In this context, we provide the comprehensive review of conventional as well as recent developments in background subtraction to analyze the success and current challenges in this field. Firstly, this paper introduces the overview of background subtraction process along with challenges and benchmark video datasets released for evaluation purpose. Then, we briefly summarize the background subtraction methods and report a comparison of the most promising state-of-the-art algorithms. Moreover, we comprehensively investigate some of the recent methods in order to find out how they have achieved their reported performances. Finally, we conclude with the shortcomings in the current developments and outline the promising research directions for background subtraction.
学科与关键词
文献加载失败,请重试
重试