I agree that as it stands now, using your current images for temporal object tracking would be quite difficult. However, I would be less worried about the number of cell per se, than the low image quality that seems to be the case here.
Based on my experience, I believe that achieving suitable object segmentation is the most challenging aspect of object tracking to overcome. Even if you have a lot of cells present, if they are well-separated and easily identifiable form one another, you can still make substantial progress. This is most often done by having a nuclei stain, or a suitable surrogate. In lieu of a nuclei stain, perhaps the dark space in the cytoplasmic stain can be used as a surrogate as you describe. However, there are two problems I see based on this image:
- The high cell density and degree of confluence/overlapping will make differentiating even the cytoplasm difficult.
- The presence of both the bright large blobs and small bright speckles will make the segmentation even harder.
I should note that these issues are general to object tracking in general, and are not specific to CellProfiler. I afraid that unless you plan to track the cells manually, the images right are just not suitable for a successful automated asssay.
Overall, I would concentrate substantial effort on optimizing the image quality to have the best chance at success with an automated solution, or even a semi-manual approach.