基于压缩感知和字典学习的背景差分法

Background Subtraction Based on Sparse Representation and Dictionary Learning

针对当使用背景差分法时,背景存在突变和渐变、图像数据的冗余和伪前景对目标检测的干扰等问题,提出一种基于稀疏表示和字典学习的背景差分法。该方法首先训练视频流得到其数据字典,并根据数据字典学习与稀疏表示理论建立背景模型,可以有效减少数据的冗余。然后根据目标及其邻域的密集度进行目标分割,以排除前景的干扰。最后再根据数据字典的更新算法,有效解决了背景的突变和渐变问题。实验结果表明,该方法具有可行性。

In this paper, we propose a CS-based background subtraction approach based on the theory of sparse representation and dictionary learning, to handle sudden and gradual background changes and the redundancy of excessive image data and the interference of prospect. This method gets their data dictionary according to the video stream and establishes the background model based on the theory of dictionary learning and sparse repre- sentation to effectively reduce data redundancy. Then, the moving objects correctly depending on the intensity of the target and its neighbors are segmented so as to rule out interference of the foreground. Finally, the problem of sudden and gradual background changes is solved through the update algorithm of data dictionary. Exper- iments show that this method is feasible.