The Research And Implementation On Multi-target Tracking In Biomedical Image Sequences

Realizing a computer assisted sperm motion analysis system is a key task in this project. So, finding a kind of rapid and precise algorithm is the key point of this project and also this paper.Kalman forecast tracking algorithm has the merits of simple recursive algorithm,small data storage and good performance in real time realizing. So comparing different kinds of tracking algorithms, the 'forecast,association,modifying' method which is based on Kalman filter is taken into account. However, the standard Kalman filter is established on the standard mathematical model which is just a kind of similitude to the real model. So there is estimation bias. With the increment of targets, the probability of superposition and motion mutation increases. That will induce the increment of estimation bias which will probably lead to the tracking failure. Based on the analysis of the motion characteristics, the retrospective method is used to compensate the bias. The system realized with this algorithm shows perfect performance in the clinical experiments.However this compensation algorithm tries to modify the bias from the results but not the essence. So, deeper researches are made based on the experiments in the realized system. A new modified algorithm is brought forward. The analysis and compensation of the estimation bias are the key points of this paper and also our innovation.The study way which is from theory to practice, then back to theory to subliming theory gives a thorough research in this subject. The successful clinical experiments testify that the settlement of this project is satisfied.