3-D Maximum Likelihood Estimation Sample Consensus for Correspondence Grouping in 3-D Plant Point Cloud

Computer vision based plant phenomics can be used to monitor the health and the growth of plants. This letter presents the extension of 2-D maximum likelihood matching to 3-D maximum likelihood estimation sample consensus (MLEASAC) and provides a comparative evaluation of some popular 3-D correspondence grouping algorithms. We test these algorithms on 3-D point clouds of plants along with two standard benchmarks addressing shape retrieval and point cloud registration scenarios. The performance of the correspondence grouping algorithms is evaluated in terms of precision and recall. The results show that of all the evaluated algorithms, 3-D random sample consensus (RANSAC) and MLEASAC perform the best, with MLEASAC being slightly more efficient while being computationally less intense than RANSAC.