This paper describes a tracking method to trace the movements of fluorescently labeled mitochondria in time-lapse image sequences. It is based on particle filtering, which is a state-of-the-art tracking method, and is enhanced with a spatial constraint to improve robustness. Since mitochondria move only through axons, the spatial constraint is generated by axon segmentation on a single frame, which is the average of all the frames. The spatial constraint limits the search space of the state vector and, consequently, lowers the chance for the tracking to get lost. Using a background subtraction algorithm, the proposed method is also equipped with automatic detection of starting points, thus minimizing the requirement for user input. Implementation of the proposed method for tracking of fluorescently labeled mitochondria in time-lapse images showed substantially improved robustness and speed compared to a conventional method. With these improvements, this new particle tracking method is expected to increase the throughput of fluorescently labeled mitochondrial transport experiments, which are required for neuroscience research.