一种自适应双码本模型目标检测算法

Target Detection Algorithm Based on Self_adaptive Dual Codebook Model

自适应码本(SACB)算法在目标物体检测中取得了较好的效果,对周期性晃动有一定抗干扰性,但仍存在"鬼影"现象、对密集型物体晃动抗噪性差、码字清除策略过分依赖于时间等问题.对此提出一种改进的自适应双码本(DECB)算法,通过增加前景像素记忆层构成背景-前景双码本模型,实时地将频繁出现的伪前景加入背景模型中,以消除"鬼影"现象;利用相邻像素的相似时空特性,随机与相邻像素码本匹配以提高对密集型物体晃动的抗噪能力;在码字更新机制中引入生命周期,以降低码字清除策略对时间的依赖性.实验结果表明,DECB算法能够有效消除"鬼影"现象,提高抗噪性且实时性良好.

Serf-adaptive Codebook （ SACB ） algorithm, which has a particular anti-interference performance on periodic shaking, has achieved good effect in the target object detection, but there still exist some drawbacks such as ＂empty shadow＂, poor noise-resistance to intensive objects shacking and codeword-clearing mechanism overdependent on time. To address these problems, an improved self-adaptive dual codebook （ DECB ） algorithm is proposed. By constructing background-foreground dual codebook model with the introduction of a foreground pixel memory layer and moving the pseudo-foreground data which appears frequently into background model in real time,the ＂empty shadow＂ is eliminated. To enhance the noise resistance of intensive objects shaking, the pixel is randomly matched with the codebook of its adjacent one, taking the advantage of the similarity of spatiotemporal characteristic among adjacent pixels. The dependence of codeword-clearing strategy on time is reduced by introducing life-style to codeword update mechanism. The experimental results show that DECB can effectively eliminate the phenomenon of ＂empty shadow＂ and achieve both higher noise resistance and detection accuracy with better real-time performance.