融合高斯混合建模和图像粒化的猪只目标检测

Detection of pig object detection by fusing Gaussians mixture modeling and image granulation

为实现圈养猪只目标检测,克服高斯混合建模(GMM)不能很好地提取缓慢运动及静止目标的缺陷,试验提出了一种融合GMM和图像粒化的运动猪只目标检测方法,采用GMM提取图像运动像素,获取粗糙前景图;提出一种图像粒化方法,得到粒化图,将单个同质区域所包含的像素点聚集为一个图像粒;依据图像序列的前景图分析粒子运动属性,融合前景图和粒化图,最终得到精确的猪只目标检测。结果表明:本方法能有效检测缓慢运动和一段时间静止的猪只目标。

To achieve captive pig object detection and overcome the flaw of Gaussian mixture modelling （GMM） that is not good extraction for slow motion and stationary object, a moving pig object detection method by fusing GMM and image granulation was proposed in the test. GMM was applied to extract the moving image pixels and obtain a rough foreground image. An image granulating method was proposed to transform all of the pixels contained in a homogeneous re,on to a granule and to obtain the granulating image. The motion attribution of granule was analyzed based on foreground image, and then an accurate pig object detection was achieved after fusing the foreground image and granulating image. The results showed that the method could effectively detect the pig objects containing slow motion and still for a period of time.