A Novel Method for Video Moving Object Detection Using Improved Independent Component Analysis

Abstract

Moving object detection is a hot and difficult problem. How to separate the changed part from the unchanged one is the key. Also, the connection between the multi-dimensional information of input frames often plays an important role in moving object detection. In this article, we presented a novel and effective video moving object detection method. The innovation of this method lies in the calculation of the connection between multi-dimensional information considering time and space domain at the same time. At present, most moving object detection algorithms consider the time dimension and the space dimension in an independent manner. It is possible to cause some detection problems, such as, if the background information of a static or low-speed object is updated too quickly, only part of the foreground region is detected correctly. To avoid such deficiencies, the improved ICA method is adopted to separate the continuous multi-frame images in the time domain, and then to integrate these signals in the space domain. It could separate and detect the difference between the input signals to extract the correct foreground target, especially in some situations that insignificant changes happen. The proposed algorithm is named VTD-FastICA. A large number of experiments have shown that our method is more competitive than most of the moving object detection methods, especially in detecting long-span, occlusion environments, and objects with weak changes. Moreover, we also analyze the efficiency of our method in different situations, indicating that the proposed algorithm can be applied to real situations.