基于双向分析的KGMM运动目标检测算法
The dynamic target detection algorithm based on bidirectional KGMM

针对传统的目标检测算法往往是顺着时间轴方向从过去到现在分析视频序列，而忽略当前帧之后的逆向视频帧信息，对于复杂场景下的背景突变或光照变化的运动目标检测等方面存在不足. 提出了基于双向分析的(KGMM)运动目标检测方法. 在KGMM模型基础上，加入向后分析建立混合高斯模型，有效解决了较强的背景扰动和环境的复杂变化带来检测效果不好的问题，提高了算法的适应性. 向前分析模型与向后分析模型共享一个高斯分布集，减少了高斯分布个数，保证了算法的运行速度. 实验结果表明，改进的算法检测效果更理想.
For most traditional target detection methods，they prefer to use information observed in past frames when analyzing the video sequences instead of the opposite direction. As for suppressing strong background disturbance and complex environmental changes of background，the performs of KGMM do not so well. In order to solve the problem effectively，we proposed a bidirectional analysis method based on KGMM model which contains a backward establishment of Gaussian mixture model. Furthermore the forward analyzing model and backward analyzing model share one Gaussian distribution set，thus our method can reduce the number of Gaussian model and improve the operating speed. Experimental results show that the improved detection algorithm performs better