摘要： 在智能视频运动目标检测中，带掩膜的背景差方法存在会出现鬼影目标和计算复杂度大等不足。为此，提出一种改进算法，利用相关矩阵判别法去除鬼影目标，同时增加对光照变化的处理，使算法快速适应光照变化。将耗时的浮点运算转化为整型运算。变换后的算法可同时适应浮点和定点处理器，更具有通用性。在DM642嵌入式系统上进行实现，结果表明，该算法速度快、检测准确率高，满足实时性要求。

关键词: 带掩膜的背景, 光照变化, 鬼影去除, 多线程缓冲池

Abstract: As there are shortcomings of ghost object and complex computation in masked background subtraction algorithm, all of which make it not so good for smart video moving objects detecting. This paper represents an ameliorative method. It can effectively distinguishing ghost objects from real objects by using correlation matrix. It also deals with illumination changes at the same time, which makes the algorithm adapt to illumination changes faster and better. In addition, it uses integer fixed point computation instead of complex float point computation without reducing precision. Thus makes the algorithm adapt to both float point processor and fixed point processor. The algorithm is realized based on DM642 embeded system and the experimental results indicate that this paper’s algorithm is fast and more credible and it can meet the real-time requirements.

Key words: masked background subtraction, illumination changes, ghost object removing, multi-thread buffer pool

中图分类号: 

• TP18