Stratified Counting Of Overlapping Particles Based On Color And Depth Images

The image processing technology of contemporary society has developed rapidly,and many automatic counting methods based on image processing have been proposed.Aiming at the research status of the current overlapping particle counting method,the accuracy of the method is not high and it is impossible to count when the particles are completely overlapped.This paper proposes a method for stratified counting of overlapping particles based on the combination of color image and depth image.Firstly,the original color image and depth image required for this article were obtained by using Microsoft's Kinect Xbox 360 depth camera(hereinafter referred to as Kinect camera).When collecting depth images,it is easy to generate random noise points and black hole areas.This paper proposes a multi-frame improved median filtering method to eliminate isolated noise points and black hole regions on depth images.In order to enhance the visual effect and facilitate observation,the depth image is subjected to gray-scale enhancement transformation.Secondly,due to the difference in position and structure of the Kinect depth camera and the color camera,there is a positional deviation between the depth image and the color image.Therefore,the Kinect needs to be calibrated and registered to eliminate the deviation of the two images.This paper uses the checkerboard method to calibrate Kinect,and uses Matlab calibration package to calculate the internal parameters,rotation and offset matrix of the two cameras.According to the distortion model of the Kinect camera,combined with the above calculation results,the registration of the depth image and the color image is realized.Finally,the stratified count of overlapping particles is completed.The improved K-means algorithm is used to extract the color image target area.According to the contour features of the target area,it is divided into three types: single,sticky and overlapping.The number of individual particles is counted by the circularity value;according to the Euler arc algorithm,the edge contour reconstruction is performed on the particles of the adhesion type and the partial overlap(the bottom layer is not completely occluded)to obtain the number of the bottom target;for the complete overlap(the bottom layer has been A completely occluded type of particle,since the outline of the completely occluded particle cannot be extracted,the average area of the individual particles reconstructed by the edge is used to obtain the numberof completely occluded particles according to the total area of the target area.In order to use the average area method to count the specific fully overlapping type particles with large deviation,the gray difference counting method is proposed to count the particles whose bottom layer is completely occluded,thereby improving the overlapping type particle counting accuracy.The above is the counting of particles in a single layer(including single and blocking types)and the underlying type of overlapping type.Since the color image does not extract the upper target of the overlapping type particles,the threshold image is segmented to complete the extraction,classification and counting of the upper target.Finally,the quantities are summarized to obtain the number of particles in the entire overlapping area,thereby achieving stratified counting of overlapping particles.In this paper,we use a paper walnut with uniform size and round shape as the counting object.In the experiment,the walnut sample particles were counted,and the method was found to accurately count the adhesion type walnuts.The average correct rate of overlapping types of walnuts was 99.53%,indicating the feasibility of the overlapping particle stratification method.This article provides new ideas for automated particle counting.