中文摘要：

针对传统混合高斯模型学习率和背景模型阈值均采用无向导式选取方法,提出了相应的自适应更新策略。学习率自适应更新算法融合图像熵和2-D学习率查找表方法,解决了光照突变时背景更新过慢及固定权值造成的不能兼顾环境适应能力和抗干扰性问题。提出一种类间最大对称交叉熵阈值自适应选取方法,弥补了传统方法在选取背景模型时仅考虑权值而未考虑决定像素点归属问题的均值和方差对分类影响这一缺陷。实验结果表明,与传统算法相比,改进算法在克服光照突变、检测准确率、实时性以及目标提取完整度上均表现良好。

英文摘要：

Aiming at the learning rate and background model threshold value of traditional Gaussian mixture model all adopted the selection method without wizard, this paper put forward the corresponding adaptive update policy. The learning rate adaptive update algorithm merge image entropy and 2-D learning rate lookup table together, resolved the problem of background update slowly when illumination changes suddenly and fixed weights could＇ t give attention to both environmental adaptation ability and anti-interference ability. It proposed an adaptive threshold value selection method based on maximum symmetric cross entropy between classes, made up for the traditional method only consider weights and didn＇ t consider the mean and variance of pixels which determined the belonging problem that impacted on the classification when choosing background model. Experiment results show that compared to traditional method,improved method performed well on overcoming the illumination mutation, detection accuracy, timeliness and the integrity of target extraction.