PROBABILISTIC FEATURE MATCHING APPLIED TO PRIMITIVE BASED REGISTRATION OF TLS DATA

Abstract

Many industrial applications require dense point clouds of the installations. Acquisition of the rooms, filled with many objects, of an industrial scene leads to many Terrestrial Laser Scanner (TLS) stations. A precise registration of all the per-station point clouds is crucial for the required accuracy of 1-2 cm of final data. Targets and tachometry, current best practice for registration, slows down the survey and limits the number of campaigns. Indoor geolocation system are faster but do not reach the final required accuracy. Otherwise, 3D primitives can be automatically extracted from the dense point clouds and possibly used for registration. In a four step primitive-based registration, Acquisition – Reconstruction – Matching – Solving, the matching is crucial. This article presents a probabilistic test for 3D lines matching using a priori distributions of approximated transformations. The stochastic model of approximated transformations and resulting uncertain lines is introduced. A test is performed on a real dataset of an industrial scene and the results are analysed. Improvements of the presented test and matching framework are also discussed.