一种融合ViBe与多特征提取的微动目标检测算法

Algorithm of Micro-motion Object Detection Based on ViBe and Multi-feature Extraction

为实现前景微动目标的准确提取,克服提取过程中的高误检率等难题,对CbCr分量、RGB和SILTP特征建立背景模型,提出一种融合多特征的ViBe背景建模改进算法。首先引入LBSP算子,改进LBP-TOP纹理编码方式,利用得到的纹理特征计算当前帧的时/空域前景概率,从而建立起接近真实背景的CbCr背景模型;然后结合局部像素复杂度和3种特征的变化情况改进ViBe判别与更新方法,利用背景减除和形态学处理得到完整的前景目标进行背景替换。实验结果表明,所提算法能有效分割视频图像中的微动目标并实现背景替换。

In order to realize the accurate extraction of micro-motion object and overcome problems like high false detection rate in the target extraction process,this paper etablished background model for CbCr component,RGB and SILTP feature,and proposed an improved algorithm of ViBe background modeling based on fusion of multi features.The algorithm improvs LBP-TOP texture operator encoding method by introducing LBSP operator,and generates the spatial and temporal domain foreground probability of a pixel by utilizing the improved LBP-TOP texture feature,which is used to gradually establish a CbCr background model close to the true background.It improves ViBe foreground determination and background updating method based on the local pixel complexity and the changing conditions of the three types of characteristics.Then,it gets the complete object detection and accomplishs the accurate background replacement of the video sequences.The experimental results show that the proposed algorithm can effectively segment the micro motionobject of the video sequences and realize the background replacement.