基于特征帧构建的运动目标检测方法

Method for moving object detection based on construction of feature frames

非参数密度估计在样本分析建模方面得到了很大的关注,尤其是核密度估计方法。但由于核密度估计方法计算量大,应用到运动目标检测方面很难达到实时效果。提出了一种特征帧构建的核密度估计方法。因为核密度估计不需要假设背景模型的密度分布函数,所有样本值又满足独立同分布的原则,所以可以通过特征帧构建的方法进行背景建模,同时应用此方法进行背景更新。实验结果表明:该方法能够适应环境变化且具有运算速度快、实时性好等特点,可以将其应用到复杂背景下的监控系统中。

Nonparameter density estimation gets great attention in sample analysis and modeling aspects, especially in kernel density estimation. Because the calculated amount of the kernel density estimation is great, it is difficult to get the real-time effects when applied to the detection of moving objects. Based on the construction of feature frames to kemel density estimation, a novel algorithm was proposed. For the kernel density estimation did not need to assume density distribution of the background model, and all the samples were in line with the principle of independence and identical distribution, background model could be constructed, and background could be updated with feature frames. Experimental results show that this algorithm based on the construction of feature frames can adapt to the change of environment and it has the characteristics of fast calculation speed and good real-time feature. It can be used in the monitoring system under complicated background.