Dictionary-based background subtraction

To detect the moving objects in the intelligence video surveillance systems, a method of background subtraction was proposed. The proposed method learned a dictionary for each position of the video by using the temporal-spatial information of the local region. Once the dictionaries were built, the information of the corresponding positions in the next frame could be expressed by the dictionaries. In this study, it was supposed that the background had a high occurrence frequency. The words of each dictionary were divided into two classes according to the hypothesis. One class was used to describe the background, while the other was employed to express the foreground. The corresponding position was judged to be the background or the foreground according to the word that had made the best contribution to the corresponding pixel. The dictionary was updated after the judgment. The proposed method was performed on the public videos against three state-of-the-art algorithms. The experimental results show the superiority of the proposed method.