Abstract: Aiming at some limits for detecting moving objects from a dynamic background，a novel method is developed to detect moving objects based on spatial-temporal information.A preliminary background model is initialized by the first frame，a binary mask produced by frame-to-frame pyramid difference and an adaptive parameter is used to update background model.Moving objects are detected in terms of background subtraction and pyramid difference information.The results show that proposed algorithm is efficient to extract moving objects and immune to noise.

Key words: Pyramid differences, background modeling, background subtraction

摘要： 针对复杂背景下的运动检测困难，提出了一种结合时域差分与尺度空间信息的运动对象检测算法。该方法首先以视频序列的第一帧作为初始背景，以相邻帧金字塔差分信息作为背景更新掩膜，并采用自适用的背景更新因子实现背景更新。在运动对象检测阶段，结合背景差信息与当前帧和背景帧金字塔差分信息，提取出运动对象。实验结果表明，该方法具有较好的实时性和自适应性，对噪声不敏感，有效提取运动对象。

关键词: 金字塔差分法, 背景建模, 背景差分