基于改进高斯混合模型的自适应前景提取

摘要

Motion foreground extraction under complicated scene is the basic part of intelligent video surveillance.Gaussian mixture,as a common background modeling method,in view of the unnecessary overhead caused by fixed number of Gaussian mixture model,a method based on the single Gaussian model as well as dynamic adjustment of Gaussian mixture model is proposed.The update rate is in real-time change according to how strongly the scene changes,which adapts well to the mutation scenario and the change of illumination.Then,the extraction of motion foreground is processed by morphological processing and the final goal is got.The experimental results show that the method is highly adaptive to background modeling,and the extracting prospects precision is improved.%在复杂场景下的运动前景提取是智能视频监控的基础部分。高斯混合模型是常用的背景建模方法，针对高斯混合模型中模型个数固化导致的无谓的系统开销，提出基于单高斯模型成长的动态个数调整形成的高斯混合模型。对模型的更新率根据场景变化的剧烈程度进行实时改变，能较好适应突发场景、光照的变化。对提取的运动前景进行形态学处理，得到最后的提取目标。实验结果表明，该方法背景建模适应性强，提取前景精度有所提升。