Counting of Visible Foreign Particles in Liquid Injection Based on Trajectory Modelling

摘要

In order to meet the requirements of counting of visible foreign particles in liquid injection, an automatic visible foreign particles counting system based on trajectory modelling is developed. The mechanical part drives containers to spin and stop abruptly and afterwards the liquid continues to spin. A sequence of images is collected. Gaussian mixture model is used to detect moving targets. In the first step we track trajectories preliminarily by nearest neighbor approach and break unreasonable associations. In the second step, we model the particle movement by damped vibration and a SVM with modeling residuals as features was used to determine whether two short candidate tracks to belong to the same trajectory. Finally, we analyze the trajectories to determine the number of visible foreign particles.