Evaluation of intravital tracking algorithms

We are exploring automated tracking for cells observed in vivo by video microscopy. In the past, three tracking algorithms have been utilized for tracking cells: the normalized cross correlation tracking algorithm, the centroid tracking algorithm and the snake based tracking algorithm. This paper focuses on the analysis, evaluation and comparison of the above three tracking algorithms using some performance measures. These performance measures include: (1) the average number of consecutive frames tracked; (2) the localization error; (3) correlation with ground truth, and (4) tracking time. One hundred in vivo (intravital) videos were used in our experiments for analysis, evaluation and comparison. The results show that snake based tracking provides the best performance, and that automated tracking of leukocytes is indeed possible.