中文摘要：

背景减法是运动目标检测的一类重要方法，它的难点在于背景动态模型的建立．非参数密度估计中的核密度估计方法是解决这一难点的十分有效的方法．但该方法的缺点是它的计算量较大，难以满足运动目标检测的实时性．针对该问题，提出一种基于关键帧的核密度估计运动目标检测算法．该算法采用提取关键帧的方式来建立背景模型，同时用此方式进行背景更新．它不仅减少了用于密度估计的样本数，而且降低了目标检测的虚警率和误检率．实验结果表明该算法能够适应环境的变化．比改进前的算法快了不止9倍，并可以有效地进行运动目标的检测．

英文摘要：

Background subtraction is a kind of important method of moving targets detection. Its difficulty lies in the establishment of the background dynamic model. As a nonparametric density estimation method, the kernel density estimation method （KDE） is an effective method to solve the difficulty. But its shortcoming is its large amount of calculation, and it is difficult to meet the requirement of real - time detection. To solve these problems, a kernel density estimation algorithm based on the key frame for moving targets detection （KDEKFD） was proposed in this paper. KDEKFD built the background model and updating the back- ground by the way of extracting key frames. It not only reduced the samples for density esti- mation, but also reduced the false alarm rate and the error detection rate of the targets detec- tion. The experiments showed that KDEKFD was suitable for the changing environment, and was no less than 9 times faster than the original KDE, which could be an effective way of moving targets detection.