基于深度图像处理技术的类圆形重叠颗粒计数

摘要

颗粒物计数的用途很广,但很费时且易出错,尤其是重叠时的颗粒物计数更加困难,而且由于技术和设备的限制,相关的研究也比较少.文中利用深度图像处理技术实现对重叠的类圆形颗粒物进行计数.首先采用Kinect相机对重叠目标采集图像,并且对采集的图像进行保存处理,将目标距离物体的实际距离与采集到的图片的灰度值进行映射,即目标与相机的距离越远,对应该点像素值越小.再对图像进行ROI提取和去噪,对图像进行预处理,求出分割阈值,将三维重叠目标计数问题转换为二维平面计数问题,简化研究问题的复杂性.文中采用高度大约为2cm、类圆形、大小均匀的砂糖橘作为研究目标.采用本文的研究思路得到目标二维平面区域后,采用Hough变换法进行圆形拟合,根据拟合结果得出重叠区域目标个数.实践表明,采用本文的方法能够满足重叠目标计数要求.%Particle count is widely used, but it is time-consuming and error prone, especially in overlapping particles count, it is more difficult, and because of technical and equipment constraint, relatively few studies have been done.In this paper, the technique of depth image processing is used to count overlapped circular particles.Firstly, using Kinect camera captures images of overlapped objects.Then making a mapping between pixel value of object images and target space distance in real world, that is the farther between target and camera,the smaller of the gary.Then the depth image preprocessing is used to extract ROI and denoising, computing the segmentation thresholds.The 3D overlap region is divided into two 2D planar regions, the 3D overlapping target counting is converted to a 2D counting problem, simplifying the complexity of research questions.In this paper, cirtus, which is about uniformly 2cm high in size, is used as the research target.The Hough transform method is used for circle fitting, and the number of overlapping area targets is obtained according to the fitting result.The practice shows that this method can meet the requirement of overlapping target counting.