If you don't have the Automated Driving System Toolbox that Mark mentioned then you can either try to find some tracking code in the Computer Vision System Toolbox, or write it yourself, or look for some open source tracking software.
To write it yourself, what you could do is first accept the labels of the first frame. Then compute all the centroids. Then if you can assume that the centroids did not move very much from one frame to the next, you can use pdist2() in the stats toolbox to find distances of every centroid in frame n+1 to the centroids in frame n. Start out by picking the two closest, and call those a match, assign the label from frame n to frame n+1 (meaning you've determined that they're the same blob), and remove them from consideration. Look at what's left and repeat, basically finding closest pairs, matching labels, and removing from consideration.
When you get down to a few blobs, you'll have to see if they are roughly the same distance away from each other as all the prios blobs. Otherwise you could get a situation where two blobs entered the field of view, and maybe some left the field of view and you'll have a hard time matching them up, so that's why you need to look at the distance and if it's way more than all the other matched distances, then they're probably different blobs, NOT a matched pair.
Writing tracking software is not easy - there are lots of weird pathological cases that need to be handled to do it properly.