Motion flow analysis in cell videos using a multi-level clustering method

Analyzing motion flow of cells is an important task for many biomedical applications. It is a challenging problem due to noise in images and uncontrolled motion of cells. In this study, a method to find regions of organized motion and direction of flow is proposed. Since dense optical flow methods might fail due to homogeneous regions and irregular motion patterns, the technique involves analyzing trajectories of strong corner features. Trajectories are clustered to find dominant flow patterns for different regions of the frame, where a multilevel clustering scheme is followed. Experiments show that the technique gives accurate results for detecting region and direction of flow.