Tracking dense crowds of moving cells with an adaptive dual ellipsoidal locator

Automated tracking of cell populations´ movement is vital for quantitative and systematic analysis of cell behaviors. However, it suffers from many challenges including complex cell morphology, undistinguished visual appearance, frequent occlusions, and irregular motion, etc. In this paper, we present a fully automatic and effective method to track hundreds of oval-shaped cells. A novel dual ellipsoidal locator is proposed firstly to recognize each individual cell in dense crowd. And it is then used in a spatio-temporal information fusion procedure to recover 2D trajectories of moving cells. To improve the performance of the proposed system, we developed a GPU based algorithm to track hundreds of cells in near real time. Experiment results show that our detection methods achieve a good performance comparable to existing state-of-the-art methods.