In this research we propose a combined cell matching and image alignment method for tracking cells based on their nuclear locations in 3D fluorescent Confocal Laser Scanning Microscopy (CLSM) image sequences. We then apply it to study the cell division pattern in the developing sepal of the small plant Arabidopsis thaliana. The method is based on geometric hashing and inherits its invariance to rotation, translation and scale. The method consists of three steps. In the first step the centroids of nuclei are detected using a previously developed cell detection algorithm, reducing the CLSM volumes to 3D point clouds, wherein every point represents a nuclear centroid with an associated confidence level. In the second step centroids between images are matched in two phases. First geometric hashing is used to find an initial set of centroid matches, then using the initial matches a dense matching is obtained through a novel iterative point matching algorithm. In the last step centroid matches are used to estimate transformations and register all input images to a common frame. Our algorithm has successfully aligned 12 volumes encompassing 72 hours data set and matched 258 nuclear lifelines.