Extraction et évaluation du réseau routier urbain à partir des images satellitaires développement d'algorithmes

Abstract

The objective of this thesis is to develop automatic tools for the extraction and the evaluation of urban road network from satellite imagery.The subject of the detection of urban road network is interesting because of it's usefulness and the diversity of it's applications (urban cartography, automatic navigation, recalage of images, etc.). Our research follows two axes. On the one hand, we have developed a road detection procedure, and on the other hand, in the perspective of conducting a result evaluation, we have implemented a method for the evaluation of extraction results.The proposed method provides a reliable solution to the evaluation problem because it combines both visual and quantitative measurements designed for the overall road network as well as for each of it's components.The road detection procedure is used to extract the road network from a panchromatic SPOT image of an urban area, while the evaluation algorithms are applied to the extracted network and another network detected by Wang, D. et al., (1996). We were thus able to evaluate the performance of the two different detection methods.The results obtained show a precision of 0.73 for our method and 0.70 for the Wang, D. method, while the error of commission reaches 0.47 for the proposed method versus 0.34 for the second method. To provide more depth to the analysis of our results, we carried out an evaluation for each component of the road network, namely: straights, curves, intersections and dead ends. Overall, dead end show minimal rate of successful followed by intersections, curves and straights.