Fitting Cylindrical Objects in 3-D Point Cloud using Contextual and Geometrical Constraints

摘要

In this paper, we propose a framework for fitting cylindrical objects toward deploying an object-finding-aided system for visually impaired people. The proposed framework consists of a RANSAC-based algorithm and a model verification scheme. The proposed robust estimator named GCSAC (Geometrical Constraint SAmple Consensus) avoids expensive computation of the RANSAC-based algorithms due to its random drawing of samples. To do this, GCSAC utilizes some geometrical constraints for selecting good samples. These constraints are raised from real scenarios or practical applications. First, the samples must ensure being consistent with the estimated model; second, the selected samples must satisfy explicit geometrical constraints of the interested objects. hi addition, the estimated model is verified by using contextual constraints, which could be raised from a certain scene such as object standing on a table plane, size of object, and so on. GCSAC's implementations are carried out for various estimation problems on the synthesized dataset. The comparisons between GCSAC and MLESAC algorithm are implemented on three public datasets in terms of accuracy of the estimated model and the computational time. Details of algorithm implementation and evaluation datasets are publicly available.