Premium
Visibility prediction based on artificial neural networks used in automatic network design
Author(s) -
Saadatseresht Mohammad,
Varshosaz Masood
Publication year - 2007
Publication title -
the photogrammetric record
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.638
H-Index - 51
eISSN - 1477-9730
pISSN - 0031-868X
DOI - 10.1111/j.1477-9730.2007.00454.x
Subject(s) - visibility , computer science , artificial neural network , artificial intelligence , task (project management) , workspace , photogrammetry , computer vision , point (geometry) , machine learning , engineering , mathematics , geometry , systems engineering , robot , physics , optics
Automatic design of photogrammetric networks is a complex task for which the visibility and quality constraints need to be both modelled and satisfied simultaneously. The task becomes even more complex when measurements are carried out for the first time on a large and/or complex object surrounded by multiple obstructions in a confined workspace. In this situation, automatic visibility prediction of a target point becomes an extremely difficult task. The visibility information inherent within the initial photogrammetric network can be used to solve this problem. However, this introduces some uncertainty into the prediction result because of the incompleteness of the visibility information. In a previous study, the authors developed an analytical deterministic method, visibility uncertainty prediction (VUP), that used ‘‘visibility spheres’’ to predict the visibility of target points. This paper investigates the use of artificial neural networks (ANNs) in visibility prediction, and presents a new technique, ANN‐based visibility uncertainty prediction (AVUP), that works by training a feed‐forward multi‐layer ANN. The visibility data for this network is extracted from the initial photogrammetric network. Once trained, the network can be used to predict the visibility of any target point from a potential camera station. Various experiments were carried out to evaluate the proposed technique. The results showed that, compared to the previous deterministic method, it is more accurate and has a lower computational cost.