A Smart Learning Tracking Machine for Tracking Multiple Fluorescent Molecules in the Presence of Photobleaching Effect

This paper studies a motion correspondence problem in which a newly arise missing and recovery issue is presented. The original independent assumption on missing detection will be invalid for the tracking problems encountered via fluorescent techniques because of the ´photobleaching´ phenomenon. We designed a statistical model based on the log likelihood with the stochastic modeling of photo bleaching to tackle the problem. Our model can correctly distinguish the missing from death and is capable to trace between the missing object and its recovery, even for a long period of missing detection. Besides, modifications to existing optimization problems are discussed and an efficient and novel algorithm to search the global optimal is proposed. Additionally, we introduce the EM algorithm with Monte Carlo approximation to realize the automatical learning of the parameters of the model. Several experiments are presented in the simulation part to show the model´s high accuracy in dealing with sophisticated scenes and thus demonstrate its merit in facing of the new challenge.