You can save the centroid results into a different variable at the end of every loop - 'centroids_old', and then compute the Euclidean distance between 'centroids_old' and 'centroids' for every row.
distCovered(k) = sqrt(sum((centroids(k,:)-centroids_old(k,:)).^2));
You will have to take care of boundary cases, where a track is lost (a row in centroids deleted) or a track is created (a row added in centroids).
