【摘要】 为了还原动态背景像素值的真实分布,本文提出了基于最优k均值聚类的时空背景模型.首先采集每个像素点不同时刻的相邻像素信息,然后对采集到的所有样本像素值进行多次k均值聚类,并计算对应的轮廓系数找到最优k值,建立初始背景概率模型.由于最优k值反映了数据真实分布的个数,其值越大说明动态背景变化越快,因此最后根据最优k值计算更新速率对背景模型进行实时更新.本文在CDnet2014提供的动态背景数据集上进行了相应的实验,实验结果表明本文提出的模型对运动目标检测的效果要优于ViBe、EFIC、AAPSA等目前已有的算法.更多还原

【Abstract】 In order to restore the real data distribution of dynamic background pixel values,the spatiotemporal background model based on optimal k-means clustering is proposed. First,the information of neighbouring pixels is collected at different time point. Then the kmeans clustering algorithm is used multiple times to cluster the sample data collected,and the Silhouette Coefficient is computed to find the the optimal k so that the initial background probability model can be established. The optimal k reflects the number of real data’s distributions,and the greater the value is,the faster the dynamic background changes. So the model is finally updated with the update rate which is computed by the optimal k. This paper has carried out a corresponding experiment on the dynamic background data set provided by CDnet2014. The experimental results showthat the proposed model is better than the existing algorithms such as ViBe,EFIC,AAPSA and so on.更多还原