Automated Assessment of Road Generalization Results by Means of an Artificial Neural Network

Abstract

This paper develops a method and tool for automatically evaluating road generalization quality when one goes from a scale of 1:25,000 to 1:100,000 and 1:200,000. The automation process forms part of the objective because it improves evaluation productivity. In order to perform that process, two conditions are necessary: (a) expert knowledge, which guarantees evaluation reliability; and (b) a tool that formalizes that knowledge. We obtained these objectives by (a) asking a group of 25 international experts to assess the quality of a representative sample of lines, and (b) implementing an artificial neural network (ANN) as a tool for automated selection of the best-generalized line. After calibrating two ANNs by training, the agreement with expert scores was over 73% and 80% at scales of 1:100,000 and 1:200,000, respectively.