An Experimental Comparison of Range Image Segmentation Algorithms
IEEE Transactions on Pattern Analysis and Machine Intelligence
Efficient message passing interface (MPI) for parallel computing on clusters of workstations
Journal of Parallel and Distributed Computing - Special issue on workstation clusters and network-based computing
Segmentation of Range Images through the Integration of Different Strate Gies
VMV '01 Proceedings of the Vision Modeling and Visualization Conference 2001
Some Further Results of Experimental Comparison of Range Image Segmentation Algorithms
ICPR '00 Proceedings of the International Conference on Pattern Recognition - Volume 4
A geometric approach to the segmentation of range images
3DIM'99 Proceedings of the 2nd international conference on 3-D digital imaging and modeling
| Hi-index | 0.00 |
We report on the implementation of a range image segmentation approach on a cluster using Message Passing Interface (MPI). The approach combines and integrates different strategies to find the best fitting planes for a set of three dimensional points. There are basically three distint modules for plane recovery; each module has a distinct method for generating a candidate plane and a distinct objective function for evaluating the candidate and selecting the "best" plane among the candidates. Parallelism can be exploited in two different ways. First, all three modules can be executed concurrently and asynchronously by distinct processes. The scheduling of the modules in the parallel implementation differs significantly from that of the sequential implementation. Thus, different output images can be obtained for the two implementations. However, the experiments conducted on several range images show that on average the quality of results is similar in comparison with ground truth images. Second, the computation within each module can be performed in parallel. A module chooses the best plane among a large set of randomly selected candidate planes; this is a highly parallel task that can be efficiently partitioned among a group of processors. The approach proposed in this paper for a multiprocessor environment has been implemented on a cluster of workstations using MPI. Preliminary results are presented.