Toolbox Adaplin para QGIS: uma ferramenta de programação dinâmica para a extração semiautomática de estradas

Abstract

The production of geographic data and the synthesis of this knowledge on maps are very important nowadays. In particular, obtaining and updating a cartographic road database are essential and have applications such as urban cadastre and planning of transport systems, support for the location of all commercial and residential establishments in the geographic space. A road normally is characterized on the map by its central axis. The fastest and most economical procedure is to trace the vector using an aerial or orbital image as the base. The advantage of this approach lies in the fact that virtually no field work is done. The vector is usually designed manually, through the creation and editing tools in an environment of Geographic Information Systems - GIS. In this context, the present work proposes a semi-automatic and interactive GIS tool for vector creation that optimizes the operator work allowing for less effort and consequently better productivity at job. This tool has been implemented as a toolbox for the open-source software QGIS. The Adaplin toolbox tool adds a pair of intermediate points for each user-supplied segment, transforming this segment into a polyline. For this task, an objective function is modeled to maximize the framing of the polyline to the road in the image. The optimization procedure models the points from a graph and produces the solution by dynamic programming. To help the operator select the next point, the tool generates a real-time preview of the polyline trace. Experiments covered the comparison of the manual restitution with the one performed through the tool. The experiments database was composed of 100 cutouts with dimensions 4 × 4 km (800 x 800 px) obtained from Rapideye orbital images, of medium resolution (distance between pixels of 5 meters). The cutouts were divided into 5 classes, each class with 20 cutouts, according to the coating material and the road surroundings. For the paved road and dirt road classes surrounded by undergrowth or undergrowth and forest, the experiments showed a reduction in restitution time ranging from 10% to 21% and a reduction in the number of mouse clicks ranging from 48% to 65%, with the positional errors being maintained at an acceptable level for mapping at 1:50,000 scale. The tool did not show good results for the paved road class surrounded by exposed soil, despite the reduction in the number of clicks needed to digitize the road. The application of the same algorithm, however using the infrared band, is glimpsed for the extraction of drainage stretches corresponding to rivers flows