Machine Vision Measuring Method For 3D Positions Of Particles In Liquid Containers Using Image Distortion Features

In the mining,chemical,food and other industrial fields,solid particles are usually transported in cylindrical pipes with using liquid as medium,forming complex two-phase flows.The movement of particles in the pipe directly affects the efficiency and life span of fluid machinery.At present,many mechanism simulation models are available for predicting particle trajectories in the liquid.The model accuracy still needs to be verified by a lot of experimental data,however.The rapid development of machine vision technology has provided a new solution for experimental measuring of particle trajectories in liquids.However,the lens effect of the cylindrical container filled with liquid often results in particle image distortion.Therefore,current machine vision techniques are only capable of measuring particle trajectories in square containers,which cannot fully reflect the actual situation.Futhermore,only two-dimensional coordinates of the particle can be measured,without the depth position of the particle.In order to sovle the problem of three-dimensional position measurement of particles in cylindrical containers filled with liquid,a new machine vision measuring method is proposed in this paper,which ultilizes the deformation features of the particle image.First,an experimental platform was built to collect a large number of particle images using one camera;then,the relationship betwee distortion degree and particle position was analyzed,and support vector machine(SVM)model was established to predict the particle depth position;finally,the proposed method was implemented using Matlab programmimg language,with which three-dimensional position of the particles in the cylindrical liquid container can be meassured.The results were compared with the real trajectory of the particles.Results of this paper provide a new solution to the 3D measurement of particle trajectories in cylindrical liquid containers.The main contributions and conclusions of the paper are as follows:(1)An experimental platform for machine vision based measurement of particle position was established.Steel balls and cylindrical glass container filled with water were ultizied as testing material.More than 2000 particle images were collected by a CCD camera for different particle diameters and container diameter,which provide good basis for theoretical study.(2)The particle images were preprocessed to extract the 2D positions of the particles and to calculate the distortion features.Data analysis shows that the distortion generally increases with increasing position depth and distance from the center axis of the cylinrical container.(3)A SVM model was established to predict the depth position of the particle,with dimensionless distortion features as model input.1200 sets of data were used to train the SVM model,while 800 sets for model testing.Results show that the established SVM model has satisfactory measuring accuracy,with average error less than 5%.It is more advantages than Kriging interpolation and Random Forest modeling methods.(4)A software was designed to realize the proposed mehtod with Matlab language,including users' interface,modules of particle image importing,particle extraction,3D coordinate measurement.Using this software,the 3D coordinates of the particles at 180 different positions in the cylindrical water container were measured and compared with the real positions of the particles.The error was found to be less than 5%,indicating the effeciveness of the proposed method.