视觉注意机制是滤除冗余，获取有效信息的主要方法之一，如何合理的与已有算法相融合是当前研究的关键。本文在对视觉注意机制与运动检测算法深入分析的基础上，针对两种算法的不足，提出了一些改进，并将其引入行人检测中。通过实验分析对改进算法的有效性和稳定性进行验证。主要工作如下： 首先，为对抗现有注意模型在颜色空间及时间信息等方面存在的不足，提出了相应的改进模型。使用更接近于人类视觉的Lab颜色空间来取代传统模型中的RGB特征。并使用“胜者取全”（Winner Takes All,WTA）的决策机制获得显著特征图。 其次，在区域检测方面，针对三帧差分检测边缘不连续，空洞现象严重等缺陷，提出将改进的三帧差法与基于区域搜索迭代的内轮廓填充法相融合，以填充空洞。融合改进的视觉注意模型，提出一种基于视觉注意机制的区域检测算法，算法融合了运动检测与视觉注意的现有研究成果，实现了对运动显著区域的检测。实验结果表明，本文算法对待检测区域的目标信息起到明显的增强作用，并一定程度抑制背景干扰，是一种有效的区域检测算法。 最后，在行人检测方面，基于视觉注意的区域检测可得到被增强的包含目标的区域，称为感兴趣区域（Region Of Interest，ROI）。针对传统行人检测算法需提取整幅图像区域的HOG特征，耗时久，且误检严重；而单纯引入运动信息，漏检和局部检测现象又十分严重等不足。本文将以上区域检测算法引入其中，提出了一种基于视觉注意的行人检测算法，该方法只需在ROI内提取HOG特征并进行匹配，在各场景下的实验结果表明，本文算法的检测结果对比其他算法准确性更高。

It is an important methods to neglect useless information and obtain useful information based on visual attention model, how to combine with the existing algorithms is the focus of current research. Based on visual attention and motion detection on the basis of indepth analysis, Aiming at the shortcomings of the two algorithms, Put forward some improvement and introduce pedestrian detection. Through the experiment to validate effectiveness and stability of the improved methed. The main work is as follows: Firstly, in order to fight against existing model edge information is not ideal,lack of time information. An improved model of visual attention is put forward. Usimg more close to human visual Lab space instead of the color features in traditional model. The last, through the winner takes all (WTA) to obtain the final saliency map. Secondly, for region detection , according to the three frame difference detection edge discontinuity, Hollow is serious.It is proposed to combine three difference improved algorithm with the inner contor filling method based on iterative search ,to fill the vold. Fusion improved visual attention model , a region detection algorithm based on visual attention is proposed, fusioning the existing research results of motion detection and visual attention,achieving the effective detection of moving significant regions. The experimental results show that, This algorithm treat detection area of the target information plays a significant enhancement effect, and background interference was reduced to a certain extent, it is an effective region detection algorithm. Finally, for pedestrian detection, through region detection based visual attention,the enhanced region containing targe is obtained, be called Region Of Interest，ROI. in view of the traditional pedestrian detection algorithm for HOG needs to extract all image region. Take a long time and serious error; and when only introducing motion information Lack of missing and detect the local phenomenon is very serious.The region detection algorithm is introduced in this paper, and a pedestrian detection algorithm based on visual attention is proposed, This method only needs extraction and match HOG feature in ROI. The experimental results show that under various scenarios, The detection effect of this algorithm in this paper is compared to other algorithms with higher accuracy.