Counting Of Circular Overlapping Particles Based On Kinect

The particle analysis based on image processing technology has attracted the attention of more and more scholars.It has been involved in the fields of agricultural production,industrial inspection,medical research and so on.And It has good research significance and application prospects.Using image knowledge to study the counting of particles is also a very important application of image technology,which can replace traditional manual counting.Not only can it improve the efficiency of particle counting,but it also ensures the reliability of statistical results.In this paper,it focuses on the counting of overlapping particles based on the Kinect depth camera.Firstly,according to the development document of Kinect,combined with the Kinect model,we choose appropriate hardware and software equipment and set up Kinect development platform.Taking C++ as a compiled language,combined with OpenCV machine vision library and VS2013 is used as a simulation platform to verify whether the image acquisition system is successfully installed.Unlike the RGB cameras,Depth camera of Kinect is based on the infrared imaging principle.In the captured image,each pixel stores the distance between the camera and the target.Combining the storage range of distance information and the distribution range of grayscale information,a custom conversion formula is used to convert distance information into gray information.Secondly,for the random noise existing in the Kinect depth image,this paper introduces three kinds of filters.We analyzes their structure,filtering principles and processing results respectively.and finally we select the bilateral filter.According to the idea of morphological Top-Hat transform,a custom edge detection operator is used to extract contour.Compared with other known edge operators(such as Sobel,Canny,Laplace,etc.),the proposed algorithm in this paper has wider application scope and more uniform contour edgesFinally,when the number of particles in an image is counted,an area-based and grayscale filter is set to classify the particles based on distribution type in the image.When using the watershed method to treat adhesion types,in order to control the over-segmentation of the watershed,the distance transformation is carried out firstly,and then performing the gray-scale normalization processing to strengthen the gray-scale difference at the location oftarget adhesion.Hough transform was used to reconstruct the target with incomplete contours.In view of the time and space redundancy of Hough transform,according to the principle of intersecting chord,we estimate the size range of the target.The detection interval is narrowed,and the time and space complexity are decreased.We design a GUI interface based on matlab to demonstrate the algorithm processing results.The experimental results show that the proposed algorithm not only satisfies the counting requirements of mostly common overlapping particles,but also can identify and count some types of full-occlusion.When using multiple sets of samples to test the algorithm proposed in this paper,the results show that this algorithm can not only accurately count the number of particles in each frame,but also the processing time of each sample is less than 3.5s,which is in line with the real-time of experimental processing.