Methodology for evaluation of roods computationally extracted from digital images

• Main Author: Maia, José Leonardo [UNESP]
• Publication Date: 2004
• Other Authors: dal Poz, Aluir Porfírio [UNESP]
• Format: Article
• Language: por
• Source: Repositório Institucional da UNESP
• Download full: http://dx.doi.org/10.22456/1807-9806.19564; https://hdl.handle.net/11449/309474
• topic: cartography; digital images; road; road extraction; road network evaluation

The evaluation of cartographic features (semi-) automatically extracted from digital images is of great importance in the context of the validation of computational algorithms for feature extraction. The general procedure to evaluate the geometrical quality of cartographic features is based on the comparison between the entities extracted via computational algorithms and the corresponding ones extracted manually, the latter being referred to as reference features. Such comparison between the two groups of features is performed in three steps, i.e.: (1) calculation of corresponding points between the extracted features and the reference features; (2) estimation of quality parameters (completeness, correctness, quality, mean error (RM) and root mean squared (RMSE)) based on the corresponding points; and (3) analysis involving the quality parameters obtained in step 2. The evaluation methodology was tested in many situations, involving different image data as well as extraction methodologies with different characteristics. The results show that the methodology enables a very detailed evaluation of the results regarding the extraction methods.