Tracking On Active Cells In Time Lapse Of Image Sequences Of Neuron Stem Cells Based On Kalman Filter

In multi-cells' tracking of time lapse image sequences imaged by optical microscope, correct prediction and tracking of active cells are still unsolved problems. Although the proportion of active cells in neuron stem cells is very small, but they are very valuable. It is a key task to set up an automatic tracking and analysis system on active cells in neuron stem cells' tracking.According to active cells tracking, this article adopt kalman filter algorithm which establishs in motion model, and its calculation simple and understanding easy. In the basis of existing kalman filter method, this article put forward three improvement. First, the improved of the movement model, the movement of cell tracking existing models used more uniform model.due to the active cells distance movement, the model has its limitations. this article adopt the acceleration model, this mode is more coincidence active cells movement characteristics. Secondly, according to the recognition for active cells generally select human-machine interaction form to get target cell, this affects the automatic performance. this article adopt segment images with gray threshold and level set method, get binary images, then compared to the neighboring frames, autoly identify the active cells according to the movement of the cells. Finally, in the cell tracking search algorithm, generally adopt Mean Shift algorithm, but Mean Shift algorithm difficult to find the match information in the next frame. This article, adopt the cost function idea, it only need to know active cell shape characteristic and the movement characteristic in previous frame, we can find the matching of the cell.Whether the regions of the active cells can be correctly predicted or not directly affects the accuracy and speed of tracking. After analysis and comparison of several common cell's tracking algorithms, we think the Kalman Filter prediction has some advantages, such as simple recursive, small data storage, quick speed, well real-time performance and so on. It is suitable for active cells tracking according to active cells movement characteristics. Tracking is based on Kalman Filtering predictions, matching and correction in image's Cartesian coordinates system. It may search target cells via minimizing their cost function of characteristics, updating their state and measurement equations. The experiment results in prediction and tracking of six active cells in three image sequences show that the algorithm can track segmented active cells accurately. And the errors between tracking estimate values and practical observation values are no more than 10 pixels, which satisify the requirement of local seaching.