结合背景差分和显著值筛选的运动检测算法

本文为降低背景干扰,提出了一种结合背景差分和显著值筛选的运动检测方法。该方法将VIBE运动检测方法看作一个前置处理过程,用于粗定位可疑目标区域。然后在这些可疑目标区域上,提出一种基于稀疏表示和组合Lasso回归的显著值计算方法,计算可疑区域图像块的显著值。最后通过滤除显著值过小的可疑区域图像块来降低背景干扰,提高运动检测的精确度指标。在Changedetection.net的数据子集上的仿真实验结果表明,该方法的召回率和精确度指标都较高,而且平均检测耗时少。

A motion detection method combined with background subtraction and salience value filtering

In order to reduce background interferences, a motion detection method combined with background subtraction and salience value filtering is presented. This method takes the motion detection method called VIBE as a pretreatment process for locating suspicious object regions. And on these regions, it presents a calculation method of salience value based on sparse representation and group Lasso regression, to calculate the salience value of image blocks in these suspicious regions. Finally, it reduces background interferences by filtering out suspicious regions with too small salience values, for improving the precision of motion detection. Simulation results on sub-dataset of Changedetection.net show that, this method has high recall and precision, meanwhile has less average detection time.