基于标定图像距离变换的目标分割方法

Object Segmentation Method Based on Calibration Distance Transform Image

在视频检测中经常会有待处理图像数据过大，图像处理系统开销大，图像噪声大，目标重叠和粘连等问题，本文提出了一种结合图像标定和标定图像距离变换的运动目标分割方法。首先，对被检测摄像机进行标定，根据标定后的世界坐标系抽样图像数据，；其次，对抽样后的视频图像序列进行背景差和帧差的前景检测；然后，根据世界坐标系的尺度作距离变换。再然后，按目标对象的实际物理尺寸，剔除噪声，选取种子点；最后，我们采用分水岭算法实现目标分割。实验表明，改进后的方法降低了图像处理系统的开销，有效降低了噪声影响，提高了粘连目标的分割检测。

The following problems is existed in Video Detection System：too large image date,high computational cost,noise,object overlapped and adhesion.this paper proposed a new moving target segmentation method based on image calibration and distance transform.First, the image data is sampled according to WCS（the world coordinate system）,which is calibrated by the Camera Calibration Toolbox on Matlab.Second, the foreground is detected with Background-Differencing and Frame-Difference.Third,The Distance Transform is obtained based on the scale of WCS.Fourth,base on actual physical coordinate to reduce noise and select seed points;Finally, Segmentation is finished by the Watershed algorithm.Experimental results show that this moving target segmentation method can lower computation cost and reduce noise,improve the segmentation of adhesion target.