A new algorithm of non-rigid cell tracking based on balancing feature matching is proposed. A probabilistic model is established and solved using Random Walks with Restart (RWR). An efficient method for bidirectional balance is presented to balance the adjacency matrix of RWR

and then a precise matching between objects under different deformations is completed. The position tracking is accomplished based on the matching results. At the same time

an automatic calibration to object is presented with the results of the feature matching. A precise object contour tracking is presented with a method for image segmentation. We prove by experiment that the method shows good performance in the conditions of high background similarity and fast moving objects when used for tracking moving cell under dynamic background in videos. More accurate localization and target contour description can be obtained compared with other methods.